JP7790301B2 - Processing condition setting support system - Google Patents

Description

The present invention relates to a processing condition setting support system.

Patent Documents 1 and 2 describe determining machining conditions using a stability limit diagram that defines the relationship between the rotational speed of the tool or workpiece and the maximum depth of cut at which chatter does not occur, in order to suppress chatter (also known as chatter vibration) during cutting. In Patent Documents 1 and 2, the stability limit diagram is generated using the dynamic characteristics of the tool or workpiece, represented by the mass (M), damping coefficient (C), and rigidity (K), as well as the specific cutting resistance. The specific cutting resistance is the value obtained by dividing the cutting resistance by the cutting cross-sectional area, i.e., the cutting resistance per unit cross-sectional area.

It is also known that the knowledge of experts is compiled into a database and used to determine processing conditions for processing equipment. For example, Patent Document 3 describes a knowledge model compiled into a database. The knowledge model is a model composed of multiple factors defined using various technical terms and the relationships between the factors. It is said that the knowledge model can be used to optimize processing conditions.

Patent Document 4 describes the application of machine learning when defining the relationships between factors in a knowledge model. In particular, as means for defining the relationships between factors, it describes a means for defining the correspondence relationship between the rank value of a reference factor and the rank value of a factor to be connected (rank-rank relationship), and a means for defining the correspondence relationship between the rank value of a reference factor and the range value of a factor to be connected (rank-range relationship). Furthermore, when a reference factor has a relationship with multiple factors, the degree of influence of multiple factors on the reference factor can be determined by setting contribution levels.

Patent No. 5862111 JP 2012-200844 A Japanese Patent Application Laid-Open No. 2020-177547 Japanese Patent Application Laid-Open No. 2021-071856

A stability limit diagram shows the relationship between the rotational speed of the tool or workpiece and the maximum depth of cut at which chatter does not occur. Machining conditions are determined using a stability limit diagram, for example, as follows. When determining machining conditions to improve cutting efficiency, one method is to increase the depth of cut. In this case, a rotational speed at which the maximum depth of cut on the stability limit diagram shows a large value is selected, and a depth of cut smaller than the maximum depth of cut corresponding to that rotational speed on the stability limit diagram is set. In this way, the determined rotational speed and depth of cut as machining conditions result in machining in the stable region on the stability limit diagram, and chatter can be suppressed.

However, while the stability limit diagram shows the relationship between rotational speed and the maximum depth of cut at which chatter does not occur, it is known that the value of that maximum depth of cut itself contains variations. For this reason, when determining the depth of cut as a machining condition, it was common to set the depth of cut to a value significantly smaller than the maximum depth of cut in the stability limit diagram. In other words, while the stability limit diagram can be used qualitatively when determining the depth of cut, it has been difficult to use it quantitatively. As such, there was room for improvement in the chatter occurrence determination using the stability limit diagram that had been generated up to now.

The present invention was made in consideration of these issues, and aims to provide a machining condition setting support system that enables more accurate representation of stability limit diagrams, thereby enabling more appropriate machining conditions to be determined.

One aspect of the present invention is a machining condition setting support system that includes a processor and evaluates or sets machining conditions including at least a cutting amount and a rotation speed when performing cutting while rotating at least one of a tool and a workpiece,
The processor:
a stability limit diagram generating unit configured to be able to generate stability limit diagrams in which the depth of cut that can be taken as the machining condition is divided into a plurality of sections, the stability limit diagrams corresponding to the divided depth of cut sections each defining a relationship between a rotation speed of the tool or the workpiece and a maximum depth of cut that does not cause chatter;
a chatter occurrence determination unit configured to determine whether chatter occurs in the case of an evaluation target cutting-in amount and an evaluation target rotation speed included in the machining conditions, using a target stability limit diagram corresponding to an evaluation target cutting-in amount among the plurality of stability limit diagrams that can be generated;
The present invention relates to a processing condition setting support system comprising:

In the above aspect, the stability limit diagram generation unit can generate stability limit diagrams corresponding to each of the cutting depth intervals when the cutting depths that can be obtained as machining conditions are divided into multiple intervals. In other words, there is not just one type of stability limit diagram, but multiple types corresponding to each cutting depth interval.

We discovered that the stability limit diagram changes depending on the cutting depth, and by identifying the relationship between the cutting depth and the stability limit diagram, we were able to generate a stability limit diagram corresponding to each divided cutting depth interval, as described above. Therefore, the stability limit diagram can be displayed with higher accuracy than before.

The chatter occurrence determination unit is configured to determine whether chatter will occur for the cutting depth to be evaluated and the rotational speed to be evaluated as machining conditions. This determination uses a target stability limit diagram corresponding to the cutting depth to be evaluated. As described above, there are multiple types of stability limit diagrams corresponding to each of the divided cutting depth sections. Therefore, for the determination, a stability limit diagram corresponding to the cutting depth section that includes the cutting depth to be evaluated is used. This stability limit diagram is then used to determine whether chatter will occur for the cutting depth to be evaluated and the rotational speed to be evaluated. In this way, chatter determination is made using a stability limit diagram corresponding to the cutting depth to be evaluated. This makes it possible to determine more appropriate machining conditions.

As described above, the above aspect makes it possible to provide a machining condition setting support system that enables more accurate representation of the stability limit diagram, thereby enabling more appropriate machining conditions to be determined.

FIG. 1 is a diagram illustrating a configuration of a processing condition setting support system according to a first embodiment. FIG. 10 is a diagram showing a cutting model. FIG. 2 is a diagram showing the configuration of a management device that constitutes the machining condition setting support system of FIG. 1. 4 is a diagram showing a stability limit diagram generated by a stability limit diagram generating unit configuring the management device of FIG. 3 . FIG. 4 is a diagram showing the configuration of a specific cutting resistance generating unit that constitutes the management device of FIG. 3. FIG. 6 is a diagram showing the configuration of a knowledge model storage unit that constitutes the specific cutting resistance generating unit of FIG. 5 . FIG. 7 is a diagram showing the concept of a conditional probability table which is a part of the knowledge model storage unit in FIG. 6. FIG. 6 is a diagram showing the configuration of a knowledge model generation unit that constitutes the specific cutting resistance generation unit of FIG. 5 . 9 is a conditional probability table generated based on the parameters of the prior distribution stored in the prior distribution storage unit that constitutes the knowledge model generation unit of FIG. 8 . 9 is a conditional probability table generated based on a posterior distribution generated by a conditional probability table generation unit constituting the knowledge model generation unit of FIG. 8 . 1 shows the procedure for updating a conditional probability table, in which the initial state of the conditional probability table is generated using only the parameters of the prior distribution. 1 shows a procedure for updating a conditional probability table, in which an updated conditional probability table is generated based on a posterior distribution. This shows the procedure for updating the conditional probability table, in which a further updated conditional probability table is generated based on a new posterior distribution. 6 is a flowchart showing a process when forward probabilistic inference is performed by a specific cutting resistance output unit that constitutes the specific cutting resistance generating unit of FIG. 5 . 6 is a flowchart showing a process when a specific cutting resistance output unit constituting the specific cutting resistance generating unit of FIG. 5 performs backward probabilistic inference. FIG. 10 is a diagram illustrating a configuration of a processing condition setting support system according to a second embodiment.

(Embodiment 1)
1. Configuration of Machining Condition Setting Support System 1 The configuration of the machining condition setting support system 1 of the first embodiment will be described with reference to Fig. 1. The machining condition setting support system 1 targets a case where cutting of a workpiece W is performed by a tool T while rotating at least one of the tool T and the workpiece W, and evaluates or sets machining conditions such as the cutting depth and rotation speed using a stability limit diagram related to the cutting.

In particular, in this embodiment, the machining condition setting support system 1 generates a stability limit diagram using the specific cutting resistance in cutting. Furthermore, the machining condition setting support system 1 utilizes a knowledge model M for estimating the specific cutting resistance in cutting. In other words, the machining condition setting support system 1 estimates the specific cutting resistance by using the knowledge model M based on the machining conditions of the cutting, the specifications of the tool T, and the specifications of the workpiece W.

As shown in Figure 1, the machining condition setting support system 1 is composed of a management device 2 that is mainly operated by an administrator, a display device 3 that is held by the administrator and workers, and multiple machining devices 4 and 5 that perform cutting. The management device 2, display device 3, and machining devices 4 and 5 form a network and are configured to be able to communicate with each other. In other words, the management device 2, display device 3, and machining devices 4 and 5 are devices that make up the network.

The management device 2 is configured with a processor 2a, a storage device 2b, a display device 2c, and an input device 2d. The processor 2a performs calculations such as generating stability limit diagrams and determining chatter, while the storage device 2b stores the mechanical compliance transfer function (hereinafter referred to as the "compliance transfer function") and the knowledge model M. The management device 2 can also function as a server that manages the processing devices 4 and 5. The processing devices 4 and 5 are, for example, lathes, machining centers, milling machines, gear cutting machines, boring machines, etc.

2. Cutting Model The cutting model will be described with reference to Fig. 2. Fig. 2 shows a model using turning as an example, in which a workpiece W rotates about a central axis L and a tool T is fed in the axial direction of the workpiece W. However, the cutting model shown in Fig. 2 is a model that can be applied to cutting in general, and is also a model that can be applied in the same way when cutting is performed by rotating the tool T.

As shown in Figure 2, when a workpiece W is cut with a tool T, cutting resistance F occurs at a cutting point P. The cutting resistance F is represented by a principal force F1, a thrust force F2, and a feed force F3. In Figure 2, the principal force F1 corresponds to the component tangential to a circle centered on the central axis L of the workpiece W, the thrust force F2 corresponds to the component normal to the cutting point P, and the feed force F3 corresponds to the component in the feed direction of the tool T (the axial direction of the workpiece W). The principal force F1, thrust force F2, and feed force F3 are mutually orthogonal. Note that cutting resistance F can also be broken down into the principal force F1, thrust force F2, and feed force F3 in cutting processes other than those using a lathe.

3. Configuration of Management Device 2 The configuration of the management device 2 will be described with reference to Fig. 3. As shown in Fig. 3, the management device 2 includes, as a processor 2a and a storage device 2b, at least a compliance transfer function storage unit 11, a specific cutting resistance generation unit 12, a stability limit diagram generation unit 13, a chatter occurrence determination unit 14, a recommended condition determination unit 15, and a teaching unit 16. Furthermore, as shown in Fig. 3, in addition to the above units 11 to 16, the management device 2 also includes a display device 2c.

The compliance transfer function storage unit 11 stores the compliance transfer function of at least one of the tool T and the workpiece W. The compliance transfer function corresponds to the dynamic characteristics of at least one of the tool T and the workpiece W. In other words, the compliance transfer function corresponds to the dynamic characteristics expressed by the mass (M), damping coefficient (C), and stiffness (K) of the tool T or the workpiece W.

More specifically, the compliance transfer function is a transfer function of the response displacement of the tool T when an excitation force is applied to the tool T, a transfer function of the response displacement of the workpiece W when an excitation force is applied to the workpiece W, and a transfer function related to both the tool T and the workpiece W. The compliance transfer function applies a frequency response function. In other words, the compliance transfer function represents a function related to the response displacement of at least one of the tool T and the workpiece W relative to the rotation frequency of at least one of the tool T and the workpiece W. The compliance transfer function can be obtained, for example, by hammering tests, simulations, theoretical calculations, etc. Note that the method of generating the compliance transfer function is well known, so a detailed explanation will be omitted. The compliance transfer function storage unit 11 is configured by the memory device 2b.

The specific cutting resistance generator 12 generates the specific cutting resistance based on the machining conditions, the specifications of the tool T, and the specifications of the workpiece W. The specific cutting resistance corresponds to the value obtained by dividing the cutting resistance F by the cutting cross-sectional area. However, it has been found that the thrust force F2 of the cutting resistance F has a significant effect on chatter. Therefore, in this embodiment, the specific cutting resistance is the value obtained by dividing the thrust force F2 (shown in Figure 2) that constitutes the cutting resistance F by the cutting cross-sectional area.

As will be described later, the specific cutting resistance generator 12 is configured to generate an estimated value of the specific cutting resistance by performing forward probabilistic inference using the conditional probability table M1 that constitutes the knowledge model M. The specific cutting resistance generator 12 can also determine the cutting conditions, the specifications of the tool T, and the specifications of the workpiece W by performing backward probabilistic inference. The specific cutting resistance generator 12 is composed of a processor 2a and a storage device 2b.

The stability limit diagram generator 13 is configured to generate a stability limit diagram for cutting. The stability limit diagram defines the relationship between the rotational speed of the tool T or workpiece W and the maximum cutting depth at which chatter does not occur. The lower region of the stability limit diagram corresponds to the stable region where chatter does not occur, and the upper region corresponds to the unstable region where chatter may occur. The stability limit diagram generator 13 is configured by the processor 2a.

When generating the stability limit diagram, the stability limit diagram generator 13 uses the compliance transfer function stored in the compliance transfer function storage unit 11 and the estimated value of the specific cutting resistance generated by the specific cutting resistance generator 12. The specific cutting resistance used here is the specific cutting resistance corresponding to the depth of cut to be evaluated as a machining condition.

In this embodiment, the stability limit diagram generator 13 divides the depth of cut that can be obtained as a machining condition into multiple sections, and is configured to be able to generate stability limit diagrams SD1, SD2, SD3, and SD4 that correspond to each of the divided depth of cut sections. In other words, the stability limit diagram generator 13 is configured to be able to generate multiple types of stability limit diagrams SD1, SD2, SD3, and SD4, rather than just one type of stability limit diagram. Then, from the multiple types of stability limit diagrams SD1, SD2, SD3, and SD4 that can be generated, the stability limit diagram generator 13 generates one target stability limit diagram that corresponds to the depth of cut that is being evaluated as a machining condition.

For example, suppose the cutting depth available as a machining condition is 0.30 mm to 0.9 mm. In this case, the divided cutting depth intervals are, for example, 0.30 to 0.50 mm, 0.50 to 0.67 mm, 0.67 to 0.80 mm, 0.80 to 0.90 mm, etc. Note that the cutting depth intervals are unevenly spaced, but they can also be equally spaced. The available cutting depths and the numerical values of the cutting depth intervals are merely examples and can be changed as appropriate depending on the machining process (e.g., rough machining, semi-finishing, finishing), or the type of machining (e.g., turning, drilling, milling, gear cutting, etc.). Details of the multiple types of stability limit diagrams SD1, SD2, SD3, and SD4 will be described later.

The chatter occurrence determination unit 14 is configured to use a target stability limit diagram (one of SD1, SD2, SD3, SD4) that corresponds to the target cutting depth as a machining condition, out of multiple stability limit diagrams SD1, SD2, SD3, SD4 that can be generated by the stability limit diagram generation unit 13, to determine whether chatter will occur for the target cutting depth and the target rotational speed included in the machining conditions. The chatter occurrence determination unit 14 is configured by the processor 2a.

The recommended condition determination unit 15 determines recommended values for machining conditions using the stability limit diagrams SD1, SD2, SD3, and SD4 generated by the stability limit diagram generation unit 13. At this time, the recommended condition determination unit 15 can determine recommended values for machining conditions taking into account the judgment results by the chatter occurrence judgment unit 14. In detail, the recommended condition determination unit 15 sets recommended values as machining conditions that are located in the lower region of the stability limit diagrams SD1, SD2, SD3, and SD4. The recommended condition determination unit 15 is configured by the processor 2a.

The teaching unit 16 teaches the recommended values of the machining conditions determined by the recommended condition determination unit 15. As a means for teaching the recommended values of the machining conditions, the teaching unit 16 displays them on the display device 2c that constitutes the management device 2, or on the display device 3. The display device 3 displays the recommended values of the machining conditions obtained from the teaching unit 16 via communication.

The teaching unit 16 also transmits recommended values for the machining conditions to the machining devices 4 and 5 via communication. In this case, the machining devices 4 and 5 can acquire the recommended values for the machining conditions taught by the teaching unit 16 and perform cutting using the acquired recommended values for the machining conditions. In other words, the machining devices 4 and 5 autonomously change the machining conditions and perform cutting based on the changed machining conditions. The machining results when cutting is performed by the machining devices 4 and 5 are transmitted as observation data to the specific cutting resistance generation unit 12. The specific cutting resistance generation unit 12 can generate and update the knowledge model M using the observation data from the machining devices 4 and 5. The teaching unit 16 is composed of the processor 2a.

4. Details of Multiple Types of Stability Limit Diagrams SD1 to SD4 The multiple types of stability limit diagrams SD1, SD2, SD3, and SD4 that can be generated by the stability limit diagram generating unit 13 will be described with reference to Fig. 4. In this embodiment, the cutting depth that can be obtained as a machining condition is assumed to be 0.30 mm to 0.9 mm. The cutting depth ranges Aa1, Aa2, Aa3, and Aa4 into which the cutting depth is divided are assumed to be 0.30 to 0.50 mm, 0.50 to 0.67 mm, 0.67 to 0.80 mm, and 0.80 to 0.90 mm, respectively.

The stability limit diagram SD1 shown in Figure 4(a) corresponds to the cutting depth range Aa1, which is 0.30 to 0.50 mm. The stability limit diagram SD2 shown in Figure 4(b) corresponds to the cutting depth range Aa2, which is 0.50 to 0.67 mm. The stability limit diagram SD3 shown in Figure 4(c) corresponds to the cutting depth range Aa3, which is 0.67 to 0.80 mm. The stability limit diagram SD4 shown in Figure 4(d) corresponds to the cutting depth range Aa4, which is 0.80 to 0.90 mm.

In general, each of the stability limit diagrams SD1 to SD4 would be generated to correspond to a specific cutting depth. However, in this embodiment, the stability limit diagrams SD1 to SD4 are generated to correspond to each of the divided cutting depth intervals Aa1 to Aa4, rather than to a specific cutting depth. The reason for this is as follows.

As described above, the stability limit diagrams SD1 to SD4 are generated using the compliance transfer function and the estimated value of the specific cutting resistance. The estimated value of the specific cutting resistance is generated based on the cutting depth and is set to a value corresponding to one of the cutting depth intervals Aa1 to Aa4 described above. Therefore, the cutting depth intervals Aa1 to Aa4, which are the basis for calculating the estimated value of the specific cutting resistance, correspond to the cutting depth intervals Aa1 to Aa4 on the stability limit diagrams SD1 to SD4. As shown in Figure 4(a), (b), (c), and (d), each of the stability limit diagrams SD1 to SD4 corresponds to a respective cutting depth interval Aa1 to Aa4. Therefore, stability limit diagrams SD1 to SD4 are generated to correspond to each of the divided cutting depth intervals Aa1 to Aa4.

In addition, in this embodiment, the cutting depth intervals Aa1 to Aa4 are unequally spaced. The reason for this is as follows. As will be explained later, it has been found that the change in specific cutting resistance increases as the cutting depth decreases and decreases as the cutting depth increases. Therefore, the cutting depth intervals Aa1 to Aa4, which are the basis for calculating the estimated value of specific cutting resistance, are set wide in the range where the change in specific cutting resistance relative to the cutting depth is small, and narrow in the range where the change in specific cutting resistance relative to the cutting depth is large. In other words, the cutting depth intervals Aa1 to Aa4 are set wider as the cutting depth decreases and narrower as the cutting depth increases.

5. Configuration of Specific Cutting Resistance Generator 12 The configuration of the specific cutting resistance generator 12 will be described with reference to Figures 5 to 7. As shown in Figure 5, the specific cutting resistance generator 12 is configured with a knowledge model generator 21, a knowledge model storage unit 22, and a specific cutting resistance output unit 23. The knowledge model generator 21 is configured with a processor 2a and a storage device 2b, the specific cutting resistance output unit 23 is configured with the processor 2a, and the knowledge model storage unit 22 is configured with the storage device 2b.

The knowledge model generation unit 21 generates the knowledge model M. The knowledge model storage unit 22 stores the knowledge model M generated by the knowledge model generation unit 21.

As shown in Figure 6, knowledge model M includes conditional probability table M1 and probability tables M2, M3, and M4 for factors A, B, and C. Conditional probability table M1 is a model for estimating specific cutting resistance. As mentioned above, chatter is significantly affected by thrust force F2 of cutting resistance F in Figure 2. Therefore, the specific cutting resistance to be estimated is the value obtained by dividing thrust force F2 of cutting resistance F by the cutting cross-sectional area.

Conditional probability table M1 defines the probability of a factor value for child factor D occurring when the factor values for at least parent factors A, B, and C are set as conditions. Here, the specific cutting resistance is affected by the cutting conditions, the specifications of tool T, and the specifications of workpiece W. In other words, the specific cutting resistance takes on different values depending on the cutting conditions, the specifications of tool T, and the specifications of workpiece W.

Therefore, in the conditional probability table M1, parent factors A, B, and C are factors related to the cutting conditions, factors related to the specifications of the tool T, and factors related to the specifications of the workpiece W. The conditional probability table M1 may also include factors other than the above factors A, B, and C as parent factors. The child factor D is the specific cutting resistance.

In other words, the conditional probability table M1, which constitutes one of the knowledge models M, has at least the cutting conditions, the specifications of the tool T, and the specifications of the workpiece W as parent factors A, B, and C, and the specific cutting resistance as child factor D. The conditional probability table M1 is a model for estimating the specific cutting resistance based on the cutting conditions, the specifications of the tool T, and the specifications of the workpiece W.

As shown in Figure 7, machining conditions as parent factors A include, for example, the depth of cut, cutting speed, and feed rate. In Figure 7, Aa is a symbol representing the depth of cut, and Ab is a symbol collectively referring to the cutting speed and feed rate. Specifications of tool T as parent factors B include, for example, the material (substance) of tool T, the shape of tool T (cutting edge angle, etc.), the presence or absence of coating and the type of coating, etc. Specifications of workpiece W as parent factors C include, for example, the material (substance) of workpiece W, the shape of workpiece W, the presence or absence of heat treatment and the type of heat treatment, etc.

Furthermore, the machining conditions, tool T specifications, and workpiece W specifications as parent factors A, B, and C are each represented by a numerical factor value. In the case of machining conditions as parent factor A, the numerical values representing the cutting depth (Aa), cutting speed (Ab), feed rate (Ab), etc. are each the factor values of parent factor A. In the case of tool T specifications as parent factor B, the physical property values representing the material of tool T, the cutting edge angle representing the shape of tool T, and physical property values representing the type of coating are each the factor values of parent factor B. In the case of workpiece W specifications as parent factor C, the same is true for tool T specifications.

Conditional probability table M1 is conceptually represented as shown in Table 1.

The conditional probability table M1 defines probabilities P1(1) to P1(8), P1(11) to P1(18), P2(1) to P2(8), P2(11) to P2(18), P3(1) to P3(8), and P3(11) to P3(18) that correspond to factor values D1, D2, and D3 of specific cutting resistance as child factor D, when conditions are set as follows: cutting depth intervals Aa1 and Aa2 as the factor value of cutting depth Aa of parent factor A; factor values Ab1 and Ab2 representing feed rate and cutting speed of parent factor A; factor values B1 and B2 of parent factor B; and factor values C1 and C2 of parent factor C. In the conditional probability table M1 shown in Table 1, the sum of the probabilities in the vertical columns is 1. For example, the sum of the probabilities P1(1), P2(1), and P3(1) is 1.

In Table 1, the cutting depth intervals Aa1 and Aa2, which are the factor values of the cutting depth Aa of parent factor A, correspond to the cutting depth intervals Aa1 and Aa2 into which the cutting depths that can be taken as machining conditions are divided in the stability limit diagram generator 13, as described above. Furthermore, the factor values Ab1-Ab2, B1-B2, C1-C2, and D1-D3 of the other factors Ab, C, and D may each be a single value or a range of values. For example, the factor values Ab1 and Ab2 of parent factor Ab may be, for example, the cutting speed value; the factor values B1 and B2 of parent factor B may be, for example, the cutting edge angle value of tool T; and the factor values C1 and C2 of parent factor C may be, for example, a physical property value representing the material of workpiece W.

For example, if the factor values of parent factors Aa, Ab, B, and C are Aa1, Ab1, B1, and C1, the probability that the factor value D1 of the specific cutting resistance, which is child factor D, corresponds to is P1(1), the probability that the factor value D2 corresponds to is P2(1), and the probability that the factor value D3 corresponds to is P3(1). For example, P1(1) is 0.667, P2(1) is 0.333, and P3(1) is 0.

As described above, knowledge model M includes the conditional probability table M1 shown in Table 1, as well as probability tables M2, M3, and M4 for factors A, B, and C shown in Tables 2 to 4. Table 2 is probability table M2 for factor A, Table 3 is probability table M3 for factor B, and Table 4 is probability table M4 for factor C.

For example, probability table M2 for factor A defines probabilities Pa1 and Pa2 corresponding to factor values A1 and A2 of factor A, as shown in Table 2. The sum of probabilities Pa1 and Pa2 is 1. Probability table M3 for factor B defines probabilities Pb1 and Pb2 corresponding to factor values B1 and B2 of factor B, as shown in Table 3. The sum of probabilities Pb1 and Pb2 is 1. Probability table M4 for factor C defines probabilities Pc1 and Pc2 corresponding to factor values C1 and C2 of factor C, as shown in Table 4. The sum of probabilities Pc1 and Pc2 is 1.

As shown in FIG. 5, the specific cutting resistance output unit 23 outputs an estimated value of the specific cutting resistance using the knowledge model M stored in the knowledge model storage unit 22. In detail, when the specific cutting resistance output unit 23 obtains the factor values of parent factors A, B, and C, it performs forward probabilistic inference using the conditional probability table M1 that constitutes the knowledge model M. Forward probabilistic inference involves estimating the specific cutting resistance, which is child factor D, based on parent factors A, B, and C that are defined as conditions.

In this way, the specific cutting resistance output unit 23 outputs an estimated value of the specific cutting resistance as child factor D. For example, the specific cutting resistance output unit 23 can output an expected value. That is, it multiplies each of the factor values D1, D2, and D3 of child factor D by the corresponding probabilities P1(1), P2(1), and P3(1), and outputs the sum of the obtained values as the specific cutting resistance. In this case, the specific cutting resistance is "D1 x P1(1) + D2 x P2(1) + D3 x P3(1)."

When the specific cutting resistance output unit 23 obtains the target factor value of child factor D, which is the target value of the specific cutting resistance, it can also perform backward probabilistic inference using the conditional probability table M1 that constitutes the knowledge model M. Backward probabilistic inference involves estimating parent factors A, B, and C, which are defined as conditions, based on the factor value of child factor D. In this case, the specific cutting resistance output unit 23 can further use the probability tables M2, M3, and M4 for each factor A, B, and C to output factor values for the cutting conditions, tool T specifications, and workpiece W specifications as parent factors.

6. Configuration of the Knowledge Model Generation Unit 21 The knowledge model generation unit 21 generates a knowledge model M as shown in FIG. 5. In particular, the knowledge model generation unit 21 generates a conditional probability table M1 that constitutes the knowledge model M. Here, the conditional probability table M1 may be set by directly inputting the values of probabilities P1(1) to P1(8), P1(11) to P1(18), P2(1) to P2(8), P2(11) to P2(18), P3(1) to P3(8), and P3(11) to P3(18). Alternatively, the conditional probability table M1 may be generated using Bayesian estimation, as described below. Furthermore, the conditional probability table M1 may be updated using Bayesian estimation.

The configuration of the knowledge model generation unit 21 will be described with reference to Figures 8 to 13. As shown in Figure 8, the knowledge model generation unit 21 includes an observation data acquisition unit 41, an interval determination unit 42, an interval editing processing unit 43, a prior distribution storage unit 44, a conditional probability table generation unit 45, a weight storage unit 46, and a model editing processing unit 47. The observation data acquisition unit 41, the interval determination unit 42, the interval editing processing unit 43, the conditional probability table generation unit 45, and the model editing processing unit 47 are configured by the processor 2a, and the prior distribution storage unit 44 and the weight storage unit 46 are configured by the storage device 2b.

The observation data acquisition unit 41 acquires observation data 41a regarding the factor values of parent factors A, B, and C and the factor value of child factor D. The observation data 41a may be data obtained by performing actual cutting work, or may be data obtained by performing a cutting work simulation. The observation data 41a may also be data obtained by performing actual cutting work, or data obtained by performing a cutting work simulation.

The observation data acquisition unit 41 is configured to be able to communicate with the processing devices 4, 5, and is therefore able to acquire observation data 41a when actual cutting is performed by the processing devices 4, 5. The observation data acquisition unit 41 can acquire observation data 41a from the processing devices 4, 5 in real time, or it can acquire a collection of observation data 41a over a predetermined period of time rather than in real time. Furthermore, the observation data acquisition unit 41 can centrally acquire observation data 41a from multiple processing devices 4, 5.

The observation data acquisition unit 41 acquires data obtained by performing actual cutting, such as the machining conditions used in the actual cutting, the specifications of the tool T, and the specifications of the workpiece W. Furthermore, when actual cutting is performed, cutting resistance data is detected using a sensor such as a cutting dynamometer installed on the processing device 4, 5, and the cutting resistance data is divided by the cutting cross-sectional area to calculate the specific cutting resistance. The observation data acquisition unit 41 then acquires the calculated specific cutting resistance.

In this way, when actual cutting processing is performed, the observation data acquisition unit 41 can acquire observation data 41a regarding the factor values of parent factors A, B, and C and the factor value of child factor D. In addition to the cutting dynamometer, the sensors installed on the processing devices 4 and 5 can also acquire the power values (current values, etc.) of each drive motor to calculate the specific cutting resistance.

Furthermore, when a cutting simulation is performed, the observation data acquisition unit 41 acquires the machining conditions, specifications of the tool T, and specifications of the workpiece W as input information for the simulation, and can also acquire the specific cutting resistance as a result of the simulation. In this way, when a cutting simulation is performed, the observation data acquisition unit 41 can acquire observation data 41a regarding the factor values of parent factors A, B, and C and the factor value of child factor D.

The interval determination unit 42 determines discrete intervals by discretizing the factor values of parent factors A, B, and C and the factor value of child factor D in the conditional probability table M1 shown in Table 1. The interval determination unit 42 can select one of multiple discrete interval determination methods. For example, the interval determination unit 42 can determine intervals in which the factor values of each of factors A, B, C, and D are evenly discretized.

The interval determination unit 42 can also determine intervals in which the factor values of each of the factors A, B, C, and D are unevenly discretized. The cutting depth intervals Aa1, Aa2, Aa3, and Aa4, in which the cutting depth factor value of the parent factor Aa is unevenly discretized, can be set to 0.30 to 0.50 mm, 0.50 to 0.67 mm, 0.67 to 0.80 mm, 0.80 to 0.90 mm, etc., as described with reference to Figure 4.

As an example of unequal discretization, the interval determination unit 42 can also determine discrete intervals based on the mutual information I(X;Y) between each random variable in each of the factors A, B, C, and D. Mutual information I(X;Y) is a quantity that represents a measure of the interdependence of two random variables. Mutual information I(X;Y) indicates how much one variable can be predicted by understanding the other. As another example of unequal discretization, the interval determination unit 42 can also determine intervals that are discretized at equal frequency.

For example, the depth of cut interval can be determined based on the mutual information I(X;Y) between the depth of cut and the specific cutting resistance. The change in specific cutting resistance increases as the depth of cut decreases and decreases as the depth of cut increases. Therefore, the depth of cut interval determined based on the mutual information I(X;Y) is set wider as the depth of cut decreases and narrower as the depth of cut increases.

The section edit processing unit 43 is configured to accept input from the operator and edit discrete sections in accordance with the operator's input. The section edit processing unit 43 allows the discrete sections determined by the section determination unit 42 to be adjusted based on the operator's input. The section edit processing unit 43 can also create discrete sections from scratch based on the operator's input.

The prior distribution storage unit 44 stores parameters α _jk of a preset prior distribution 44a in order to determine the conditional probability table M1. For ease of explanation, the relationship between the parent factor A and the child factor D will be described below. The prior distribution 44a uses a Dirichlet distribution, as expressed in equation (1). In equation (1), j is a discrete interval of the parent factor A, and k is a discrete interval of the child factor D.

9, the prior distribution storage unit 44 stores a conditional probability table 44b in which the expected value θa _jk of the prior distribution 44a generated based on the parameter α _jk of the prior distribution 44a is used as each element value. Here, the conditional probability table 44b stored in the prior distribution storage unit 44 is generated based on the discrete interval determined by the interval determination unit 42.

The expected value θa _jk of the prior distribution 44a is expressed by equation (2): In equation (2), Na _j is the sum of the parameters α _jk of the prior distribution 44a in the discrete interval j, and Na _jk is the parameter α _jk of the prior distribution 44a in the discrete intervals j and k.

9 is a conditional probability table into which the corresponding expected value θa _jk is input for each discrete interval. Here, the parameter α _jk of the prior distribution 44a is set based on, for example, the worker's experience, information in a similar field, etc. In this case, the conditional probability table 44b based on the prior distribution 44a is a conditional probability table generated based on the worker's experience, information in a similar field, etc.

8, the conditional probability table generation unit 45 acquires the observed data 41a acquired by the observed data acquisition unit 41. The data frequency β _jk of the observed data 41a is expressed by equation (3). The data frequency β _jk of the observed data 41a is expressed by a multinomial distribution. The maximum likelihood estimator φ _jk for the data frequency β _jk of the observed data 41a is expressed by equation (4). N _j is the total number of data points of the observed data 41a in the discrete interval j, and N _jk is the data frequency β _jk of the observed data 41a in the discrete intervals j and k.

The conditional probability table generation unit 45 calculates a posterior distribution 45a using Bayesian estimation based on the observation data 41a acquired by the observation data acquisition unit 41 and the parameters α _jk of the prior distribution 44a stored in the prior distribution storage unit 44. The posterior distribution 45a is a Dirichlet distribution configured with the same discrete intervals as the prior distribution 44a.

Furthermore, based on the calculated posterior distribution 45a, the conditional probability table generation unit 45 generates a conditional probability table M1 in which the expected value θb _jk of the posterior distribution 45a is used as each element value. Here, the conditional probability table M1 is generated based on the discrete intervals determined by the interval determination unit 42. In other words, the conditional probability table M1 generated by the conditional probability table generation unit 45 is composed of the same discrete intervals as the conditional probability table 44b generated based on the prior distribution 44a. The generated conditional probability table M1 is shown in the upper part of Figure 10. In the conditional probability table M1, the value θb _jk is stored as the element value of each discrete interval.

The lower part of Fig. 10 shows a diagram for explaining in detail the discrete interval j of the parent factor A in the upper part of Fig. 10. Each element value θb _jk of the conditional probability table M1 in the upper part of Fig. 10 coincides with the expected value θb _jk of the posterior distribution 45a in the lower part of Fig. 10.

The expected value θb _jk of the posterior distribution 45a is expressed by equation (5). In equation (5), W _j is a weight. Na _jk is the value of the parameter α _jk of the prior distribution 44a in the discrete interval j. Na _j is the sum of the parameters α _jk of the prior distribution 44a in the discrete interval j. N _jk is the value of the data frequency β _jk of the observed data 41a in the discrete interval j. N _j is the total number of data pieces of the observed data 41a in the discrete interval j.

That is, the conditional probability table generation unit 45 calculates the posterior distribution 45a using Bayesian estimation based on the observed data 41a and the parameters α _jk of the prior distribution 44a, and generates a conditional probability table M1 in which the expected values θb _jk of the posterior distribution 45a are the element values.

Here, as shown in equation (5), the conditional probability table generation unit 45 does not simply perform Bayesian estimation but also takes into account the weight _Wj . In other words, the posterior distribution 45a is calculated using Bayesian estimation based on the observed data 41a and the parameter _αjk of the prior distribution 44a that takes into account the weight _Wj . Furthermore, each element value _Nbjk obtained in the process of generating the posterior distribution 45a by the conditional probability table generation unit 45 is expressed by equation (6). The element value _Nbjk corresponds to the numerator of equation (5).

For comparison, the expected value θb _jk of the posterior distribution 45a when the weight W _j is not applied is expressed by equation (7).

In this embodiment, as shown in equation (5), the expected value θb _jk of the posterior distribution 45a is calculated using the weight W _j . The expected value θb _jk of the posterior distribution 45a in this embodiment is calculated by multiplying Na _jk and Na _j , which represent the prior distribution 44a, by the weight W _j . In other words, the weight W _j represents the influence rate of the parameter α _jk of the prior distribution 44a on the observed data 41a. According to equation (5), the larger the weight W _j , the higher the influence rate of the parameter α _jk of the prior distribution 44a on the observed data 41a, and the smaller the weight W _j , the lower the influence rate of the parameter α _jk of the prior distribution 44a on the observed data 41a.

The weight _Wj is expressed by equation (8). In equation (8), _Kj is a coefficient for expressing the weight _Wj . _Naj is the sum of the parameters _αjk of the prior distribution 44a in the discrete interval j. _Nj is the total number of data in the observed data 41a in the discrete interval j. In other words, the weight _Wj is expressed by _Naj and _Nj .

In addition, when the amount of observation data 41a is small, it is preferable that the conditional probability table generation unit 45 uses Bayesian estimation based on the observation data 41a and the parameter α _jk of the prior distribution 44a to generate the conditional probability table M1, with the expected value θb _jk of the posterior distribution 45a as each element value. If a large amount of observation data 41a is obtained, the conditional probability table generation unit 45 may generate the conditional probability table M1 based only on the observation data 41a. Therefore, the conditional probability table generation unit 45 is configured to be able to select between generating the conditional probability table M1 using Bayesian estimation and generating the conditional probability table M1 using maximum likelihood estimation based on the observation data 41a.

8, the weight storage unit 46 stores the weights _Wj described above. The weights _Wj can be set to different values for each discrete interval of the factor value of the parent factor A, or can be set to the same value. As described above, according to equation (5), the larger the weight _Wj , the higher the influence rate of the parameter _αjk of the prior distribution 44a on the observed data 41a, and the smaller the weight _Wj , the lower the influence rate of the parameter _αjk of the prior distribution 44a on the observed data 41a. For example, the weight storage unit 46 stores multiple types of weights _{Wj_L} , _{Wj_M} , and _{Wj_S} as shown in equation (9).

The meaning of the weight _Wj will be explained in detail using specific numerical values. As shown in the above formula (7), the explanation will be given using the expected value _θbjk of the posterior distribution 45a that does not use the weight _Wj . Here, when _Nj4 is "80" and _Naj is "100", a comparison will be made between the case where the parameters _αjk of the prior distribution 44a are {1, 2, 2, 5} and the case where they are {100, 200, 200, 500}. The ratio of the parameters _αjk of the prior distribution 44a is {1:2:2:5} in both cases.

When the parameter α _jk of the prior distribution 44a is {1, 2, 2, 5} (when α _j4 = 5), the expected value θb _j4 of the posterior distribution 45a is "0.7727" as shown in formula (10). The calculation formula is "(5 + 80) / (10 + 100)". When the parameter α _jk of the prior distribution 44a is {100, 200, 200, 500} (when α _j4 = 500), the expected value θb _j4 of the posterior distribution 45a is "0.5272" as shown in formula (11). The calculation formula is "(500 + 80) / (1000 + 100)".

Thus, even if the ratio of the parameters α _jk of the prior distribution 44a is the same {1:2:2:5}, the expected value θb _j4 of the posterior distribution 45a will be different depending on the magnitude of the absolute value of the parameter α _jk of the prior distribution 44a. As shown in equation (10), the smaller the absolute value of the parameter α _jk of the prior distribution 44a, the closer the expected value θb _j4 of the posterior distribution 45a is to "0.8 (= 80/100)," which is the maximum likelihood estimator φ _j4 for the data frequency β _jk of the observed data 41a. On the other hand, as shown in equation (11), the larger the absolute value of the parameter α _jk of the prior distribution 44a, the closer the expected value θb _j4 of the posterior distribution 45a is to "0.5 (= 5/10)," which is the expected value θa _j4 of the prior distribution 44a.

The weight W _j can be regarded as the confidence level of the prior distribution 44a. In other words, if the confidence level of the prior distribution 44a is high, it is advisable to set the weight W _j to a large value. In this case, the expected value θb _j4 of the posterior distribution 45a will be close to the expected value θa _j4 of the prior distribution 44a. On the other hand, if the confidence level of the prior distribution 44a is low, it is advisable to set the weight W _j to a small value. In this case, the expected value θb _j4 of the posterior distribution 45a will be close to the maximum likelihood estimator φ _j4 for the data frequency β _jk of the observed data 41a. In this way, by adjusting the weight W _j according to the confidence level of the prior distribution 44a, it is possible to set the expected value θb _j4 of the posterior distribution 45a to a desired value.

In Figure 8, the knowledge model storage unit 22 stores the conditional probability table M1 generated by the conditional probability table generation unit 45 as a knowledge model M representing the relationship between parent factor A and child factor D.

The model editing processing unit 47 is configured to accept input from the operator and to be able to edit the conditional probability table M1 that constitutes the knowledge model M stored in the knowledge model storage unit 22 in accordance with the operator's input. The model editing processing unit 47 makes it possible to adjust each element value of the conditional probability table M1 shown in the upper part of Figure 10 based on operator input. Note that the model editing processing unit 47 can also create the conditional probability table M1 from scratch based on operator input.

Here, in addition to the above processing, when the observation data acquisition unit 41 acquires new observation data 41a, the conditional probability table generation unit 45 can generate a new conditional probability table M1 and update the knowledge model M stored in the knowledge model storage unit 22.

In this case, the prior distribution storage unit 44 can store each element value Nb _jk (expressed by equation (6)) obtained in the process of generating the posterior distribution 45a by the conditional probability table generation unit 45 as the parameter α _jk of the next prior distribution 44a. Then, the conditional probability table generation unit 45 can calculate a new posterior distribution 45a based on the parameter α _jk of the prior distribution 44a newly stored in the prior distribution storage unit 44 and the observation data 41a newly acquired by the observation data acquisition unit 41. Based on the newly calculated posterior distribution 45a, the conditional probability table generation unit 45 can generate a conditional probability table M1 in which the expected value θb _jk of the posterior distribution 45a is used as each element value.

11 to 13, the procedure for updating the conditional probability table M1 constituting the knowledge model M will be described in detail. As shown in FIG. 11, when no observed data 41a exists, the conditional probability table M1 is generated only by the parameters α _jk of the prior distribution 44a. In other words, each element value of the conditional probability table M1 here is the expected value θa _jk of the prior distribution 44a itself. At this point, the conditional probability table M1 constituting the knowledge model M stored in the knowledge model storage unit 22 is defined by the conditional probability table M1 (44b) expressed only by the parameters α _jk of the prior distribution 44a.

Next, assume that observed data 41a is acquired. As shown in FIG. 12 , the conditional probability table generation unit 45 calculates the parameter γ _jk of the posterior distribution 45a using Bayesian estimation based on the parameter α _jk of the prior distribution 44a and the observed data 41a, and generates the expected value θb _jk of the posterior distribution 45a. Then, the conditional probability table generation unit 45 generates a conditional probability table M1 in which the expected value θb _jk of the posterior distribution 45a is used as each element value. Then, the conditional probability table generation unit 45 updates the conditional probability table M1 constituting the knowledge model M stored in the knowledge model storage unit 22.

Next, the conditional probability table generation unit 45 stores each element value Nb _jk obtained in the process of generating the posterior distribution 45a in the prior distribution storage unit 44 as the parameter α _jk of the new prior distribution 44a. That is, as shown in FIG. 13 , the parameter α _jk of the new prior distribution 44a becomes each element value Nb _jk obtained in the process of generating the previous posterior distribution 45a. Then, assume that new observed data 41a is acquired. The conditional probability table generation unit 45 calculates a new posterior distribution 45a based on the parameter α _jk of the updated new prior distribution 44a and the new observed data 41a, and generates an expected value θb' _jk of the new posterior distribution 45a. Then, the conditional probability table generation unit 45 generates a conditional probability table M1 in which the expected value θb' _jk of the new posterior distribution 45a is used as each element value. Then, the conditional probability table generating unit 45 updates again the conditional probability table M1 constituting the knowledge model M stored in the knowledge model storage unit 22.

7. Configuration of Specific Cutting Resistance Output Unit 23 As shown in Fig. 5, the specific cutting resistance output unit 23 performs forward probabilistic inference using the conditional probability table M1 that constitutes the knowledge model M, thereby outputting an estimated value of the specific cutting resistance. The specific cutting resistance output unit 23 also performs backward probabilistic inference using the knowledge model M, thereby outputting factor values for the machining conditions of the cutting work, the specifications of the tool T, and the specifications of the workpiece W. Each process performed by the specific cutting resistance output unit 23 will be described in detail below.

The processing performed by the specific cutting resistance output unit 23 when performing forward probabilistic inference will be described with reference to Figure 14. The specific cutting resistance output unit 23 acquires factor values for parent factors A, B, and C (factor value acquisition step S1).

The specific cutting resistance output unit 23 then performs forward probabilistic inference using the conditional probability table M1 constituting the knowledge model M to obtain the probabilities P1(1) to P1(8), P1(11) to P1(18), P2(1) to P2(8), P2(11) to P2(18), P3(1) to P3(8), and P3(11) to P3(18) (shown in Table 1) of the factor values of child factor D corresponding to the obtained factor values of parent factors A, B, and C (probability acquisition step S2). For example, if the factor values of parent factors A, B, and C are Aa1, Ab1, B1, and C1, the probability P1(1) of factor value D1 of child factor D, the probability P2(1) of factor value D2 of child factor D, and the probability P3(1) of factor value D3 of child factor D are obtained.

The specific cutting resistance output unit 23 then generates and outputs an estimate of the specific cutting resistance based on the acquired probabilities P1(1), P2(1), and P3(1) (output step S3). For example, the specific cutting resistance output unit 23 can use the expected value obtained from the probabilities P1(1), P2(1), and P3(1) of the factor values D1, D2, and D3 of child factor D as the estimate of the specific cutting resistance. The specific cutting resistance output unit 23 can also use the factor value of child factor D with the highest probability as the estimate of the specific cutting resistance. The specific cutting resistance output unit 23 can also calculate the estimate of the specific cutting resistance based on the factor values of a predetermined number of child factors D in descending order of probability.

In this way, the specific cutting resistance output unit 23 can output an estimated value of the specific cutting resistance corresponding to the machining conditions (cutting depth, cutting speed, feed rate, etc.) as the input parent factors A, B, and C, the specifications of the tool T, and the specifications of the workpiece W.

Furthermore, the processing performed by the specific cutting resistance output unit 23 when performing backward probabilistic inference will be described with reference to Figure 15. The specific cutting resistance output unit 23 acquires the target factor value of the child factor D, which is the target value of the specific cutting resistance (factor value acquisition step S11).

The specific cutting resistance output unit 23 then performs backward probabilistic inference using the conditional probability table M1 that constitutes the knowledge model M (probability acquisition process S12). In the backward probabilistic inference, the specific cutting resistance output unit 23 further uses the probability tables M2, M3, and M4 for each of the factors A, B, and C. Then, using the conditional probability table M1 and each of the probability tables M2, M3, and M4, the specific cutting resistance output unit 23 can obtain the probability of each factor value of the parent factors A, B, and C that correspond to the target factor value of the specific cutting resistance.

The specific cutting resistance output unit 23 then generates and outputs the factor values of the parent factors A, B, and C, i.e., the factor values for the cutting conditions, the specifications of the tool T, and the specifications of the workpiece W, based on the calculated probability of each factor value of the parent factors A, B, and C (output step S13). In this way, the specific cutting resistance output unit 23 can output the factor values of the parent factors A, B, and C that correspond to the input target factor value of the specific cutting resistance, i.e., the factor values for the cutting conditions, the specifications of the tool T, and the specifications of the workpiece W.

8. Processing of the management device 2 when focusing on the evaluation target cutting depth The processing of the management device 2 shown in Figure 3, particularly the specific cutting resistance generating unit 12, the stability limit diagram generating unit 13, and the chatter occurrence determining unit 14, will be explained with a focus on the evaluation target cutting depth. The evaluation target cutting depth is the cutting depth that is to be evaluated. In the following, an example will be explained in which the evaluation target cutting depth is, for example, 0.55 mm.

The specific cutting resistance generator 12 generates an estimated value of the specific cutting resistance using the conditional probability table M1 shown in Table 1. Here, the evaluation target cutting depth of 0.55 mm is included in the cutting depth range Aa2, which is 0.5 to 0.67 mm. Therefore, in the conditional probability table M1 shown in Table 1, it corresponds to the cutting depth range Aa2, which is the factor value of the cutting depth Aa of the parent factor A. Then, using the conditional probability table M1, the specific cutting resistance generator 12 generates an estimated value of the specific cutting resistance D, which is a child factor, corresponding to the cutting depth range Aa2 of the cutting depth, which is the parent factor Aa, and the factor values of the other parent factors Ab, B, and C.

Next, the stability limit diagram generator 13 generates a stability limit diagram using the compliance transfer function stored in the compliance transfer function storage unit 11 and the specific cutting resistance generated by the specific cutting resistance generator 12. The stability limit diagram generator 13 generates stability limit diagrams SD1 to SD4 corresponding to the cutting depth intervals Aa1 to Aa4. In this example, the cutting depth to be evaluated is 0.55 mm, so the stability limit diagram SD2 corresponding to the cutting depth interval Aa2 is generated.

The chatter occurrence determination unit 14 then uses the stability limit diagram SD2 generated by the stability limit diagram generation unit 13 to determine whether chatter will occur for the evaluation target cutting amount and evaluation target rotational speed. In other words, the chatter occurrence determination unit 14 determines that chatter will not occur if the position corresponding to the evaluation target cutting amount and evaluation target rotational speed is in the lower region of the generated stability limit diagram SD2.

9. Effects According to the above-described machining condition setting support system 1, the stability limit diagram generating unit 13 can generate the stability limit diagrams SD1 to SD4 corresponding to each of the cutting-in amount sections Aa1 to Aa4 when the cutting-in amount that can be taken as a machining condition is divided into a plurality of sections. In other words, there is not one type of stability limit diagram SD1 to SD4, but multiple types corresponding to each of the cutting-in amount sections Aa1 to Aa4.

We discovered that the stability limit diagrams SD1 to SD4 behave differently depending on the cutting depth, and by identifying the relationship between the cutting depth and the stability limit diagrams SD1 to SD4, we were able to generate the stability limit diagrams SD1 to SD4 for each of the divided cutting depth intervals Aa1 to Aa4, as described above. Therefore, the stability limit diagrams SD1 to SD4 can be displayed with greater accuracy than previously possible.

The chatter occurrence determination unit 14 is configured to determine whether chatter will occur for the evaluation target depth of cut and evaluation target rotational speed as machining conditions. This determination uses the target stability limit diagram SD1-SD4 corresponding to the evaluation target depth of cut. As described above, there are multiple types of stability limit diagrams SD1-SD4 corresponding to each of the divided depth of cut intervals Aa1-Aa4. Therefore, the determination uses the stability limit diagram SD1-SD4 corresponding to the depth of cut interval Aa1-Aa4 that includes the evaluation target depth of cut. The stability limit diagram SD1-SD4 is then used to determine whether chatter will occur for the evaluation target depth of cut and evaluation target rotational speed. In this way, chatter determination is performed using the stability limit diagram SD1-SD4 corresponding to the evaluation target depth of cut. This makes it possible to determine more appropriate machining conditions.

The machining condition setting support system 1 also includes a processor 2a and a storage device 2b. The storage device 2b includes a compliance transfer function storage unit 11 that stores a compliance transfer function, which is a transfer function of the response displacement when an excitation force is applied to at least one of the tool T and the workpiece W. The processor 2a also includes a specific cutting resistance output unit 23 that is configured to output an estimated value of the specific cutting resistance, which is the value obtained by dividing the cutting resistance by the cutting cross-sectional area, and which corresponds to each of the divided cutting depth intervals Aa1 to Aa4, based at least on the cutting depth.

The stability limit diagram generator 13 is configured to generate target stability limit diagrams SD1 to SD4 corresponding to the depth of cut intervals Aa1 to Aa4 that include the target depth of cut, based on the compliance transfer function and an estimated value of the target specific cutting resistance corresponding to the target depth of cut as a machining condition. The chatter occurrence determiner 14 is configured to determine whether chatter will occur using the target stability limit diagrams SD1 to SD4 generated by the stability limit diagram generator 13.

In this way, the stability limit diagrams SD1 to SD4 are generated using the compliance transfer function and the estimated value of the specific cutting resistance. The estimated value of the specific cutting resistance corresponds to the depth of cut to be evaluated. Therefore, the generated stability limit diagrams SD1 to SD4 are more accurate.

The storage device 2b constituting the machining condition setting support system 1 further includes a knowledge model storage unit 22 that stores the knowledge model M. The knowledge model M is configured to include a conditional probability table M1 in which the parent factors A, B, and C are at least the cutting depth, the specifications of the tool T, and the specifications of the workpiece W among the machining conditions, and the specific cutting resistance is the child factor D, and in which the probabilities P1(1) to P1(8), P1(11) to P1(18), P2(1) to P2(8), P2(11) to P2(18), P3(1) to P3(8), and P3(11) to P3(18) that correspond to the factor value of the child factor D when the factor values of the parent factors A, B, and C are used as conditions are defined.

The specific cutting resistance output unit 23 is configured to output an estimated value of the specific cutting resistance as child factor D by performing forward probabilistic inference using the conditional probability table M1 when it obtains the factor values of parent factors A, B, and C.

In this way, in the machining condition setting support system 1, the conditional probability table M1 is configured as the knowledge model M. The conditional probability table M1 is defined by the probability of the factor value of child factor D occurring when the factor values of parent factors A, B, and C are used as conditions. In other words, the conditional probability table M1 does not directly relate the factor values of parent factors A, B, and C and child factor D, but rather represents them using probabilities P1(1) to P1(8), P1(11) to P1(18), P2(1) to P2(8), P2(11) to P2(18), P3(1) to P3(8), and P3(11) to P3(18). This allows for flexible definition of the relationships between factors.

Parent factors A, B, and C include the cutting conditions, the specifications of tool T, and the specifications of workpiece W. Child factor D is the specific cutting resistance. By using conditional probability table M1 as knowledge model M, even if there are many different types of parent factors A, B, and C, it becomes easy to define the probabilities P1(1) to P1(8), P1(11) to P1(18), P2(1) to P2(8), P2(11) to P2(18), P3(1) to P3(8), and P3(11) to P3(18) that correspond to the factor value of child factor D, which is the specific cutting resistance, when the factor values of these multiple parent factors A, B, and C are used as conditions.

Then, when the specific cutting resistance output unit 23 acquires factor values related to the machining conditions, tool T specifications, and workpiece W specifications as parent factors A, B, and C, it can perform forward probabilistic inference using the conditional probability table M1 that constitutes the knowledge model M to obtain an estimated value of the specific cutting resistance as child factor D.

Therefore, the machining condition setting support system 1 makes it possible to estimate specific cutting resistance with high accuracy by constructing and utilizing a knowledge model M related to specific cutting resistance.

In addition, the conditional probability table M1 discretizes the cutting depth as the parent factor A into each of the divided cutting depth intervals Aa1 to Aa4. Each cutting depth interval Aa1 to Aa4 is set wide in the range where the change in specific cutting resistance relative to the cutting depth is small, and narrow in the range where the change in specific cutting resistance relative to the cutting depth is large.

In particular, the change in specific cutting resistance increases as the cutting depth decreases and decreases as the cutting depth increases. Each cutting depth interval Aa1-Aa4 is set wider as the cutting depth decreases and narrower as the cutting depth increases. Therefore, stability limit diagrams SD1-SD4 can be generated with high accuracy.

The cutting resistance F that occurs during cutting is expressed by the principal force F1, the thrust force F2, and the feed force F3. In the machining condition setting support system 1, the specific cutting resistance corresponds to the value obtained by dividing the thrust force F2 that constitutes the cutting resistance F by the cutting cross-sectional area. In this way, by setting the specific cutting resistance to a value related to the thrust force F2, which has a significant effect on chatter, the specific cutting resistance related to chatter can be derived with high accuracy.

The conditional probability table M1 constituting the knowledge model M is generated using Bayesian estimation. That is, the storage device 2b constituting the machining condition setting support system 1 includes a prior distribution storage unit 44 that stores parameters α _jk of a prior distribution 44a that are preset to determine the conditional probability table M1. The processor 2a constituting the machining condition setting support system 1 also includes an observation data acquisition unit 41 that acquires observation data 41a obtained by performing actual cutting or a cutting simulation, the observation data 41a being about the factor values of parent factors A, B, and C and the factor value of child factor D, and a conditional probability table generation unit 45 that calculates a posterior distribution 45a using Bayesian estimation based on the parameters α _jk of the prior distribution 44a and the observation data 41a, and generates a conditional probability table M1 whose element values are the expected values θb _jk of the posterior distribution 45a. The knowledge model storage unit 22 stores the knowledge model M configured using the conditional probability table M1 generated by the conditional probability table generation unit 45.

In this way, by using Bayesian estimation in generating the conditional probability table M1, the prior distribution 44a and the observed data 41a can be effectively utilized. In other words, by combining the prior distribution 44a, which serves as pre-established knowledge, with the observed data 41a, the accuracy of the conditional probability table M1 can be improved.

The storage device 2b constituting the machining condition setting support system 1 also includes a weight storage unit 46 for storing weights _Wj representing the influence ratio of the parameter _αjk of the prior distribution 44a on the observed data 41a in Bayesian estimation. The conditional probability table generation unit 45 calculates a posterior distribution 45a using Bayesian estimation based on the observed data 41a and the parameter _αjk of the prior distribution 44a taking the weights _Wj into account, and generates a conditional probability table M1 in which the expected value _θbjk of the posterior distribution 45a is used as each element value.

When generating the conditional probability table M1 using Bayesian estimation, it may not be possible to obtain a high accuracy of the conditional probability table M1 depending on the accuracy of the prior distribution 44 a. Therefore, the conditional probability table generation unit 45 can generate the conditional probability table M1 according to the confidence level of the prior distribution 44 a by calculating the expected value θb _jk of the posterior distribution 45 a using the weight W _j .

Then, when a large amount of observation data 41a is secured by the conditional probability table generation unit 45 using Bayesian estimation to generate the conditional probability table M1, it is possible to increase the influence ratio of the observation data 41a in the expected value θb _jk of the posterior distribution 45a. Furthermore, when the conditional probability table generation unit 45 uses Bayesian estimation to sequentially update the conditional probability table M1, it becomes possible to secure a large amount of observation data 41a, and as a result, the influence ratio of the observation data 41a in the expected value θb _jk of the posterior distribution 45a increases. Therefore, when a large amount of observation data 41a is secured, it is possible to generate a highly accurate conditional probability table M1 by increasing the influence ratio of highly reliable observation data 41a.

The weight _Wj can be set to a different value for each factor value of the parent factors A, B, and C, i.e., for each discrete interval j of the parent factors A, B, and C. The weight _Wj represents the confidence of the prior distribution 44a. The confidence may differ depending on the discrete interval j of the parent factors A, B, and C. For example, in a discrete interval j that the operator has had many experiences with, the prior distribution 44a can be determined with a high confidence. However, in other cases, the prior distribution 44a may not be determined with a high confidence. In such cases, by setting the weight _Wj for each discrete interval j, an appropriate conditional probability table M1 can be generated.

Furthermore, the processor 2a constituting the machining condition setting support system 1 includes a recommended condition determination unit 15 that determines recommended values for machining conditions using a stability limit diagram, and an instruction unit 16 that instructs the recommended values for machining conditions determined by the recommended condition determination unit 15. In this way, by instructing the recommended values for machining conditions, workers and managers can grasp the appropriate machining conditions.

Furthermore, the processing devices 4 and 5 can be configured to be able to communicate with the processor 2a, acquire recommended values for the processing conditions taught by the teaching unit 16, and perform cutting processing using the acquired recommended values for the processing conditions. In this way, the processing devices 4 and 5 can autonomously perform cutting processing using the taught processing conditions, thereby achieving the desired cutting processing without human intervention.

(Embodiment 2)
The configuration of a machining condition setting support system 100 according to the second embodiment will be described with reference to Fig. 16. In Fig. 16, the same components as those in the first embodiment are denoted by the same reference numerals.

As shown in FIG. 16, the machining condition setting support system 100 is configured by incorporating the processor 2a and storage device 2b that constitute the management device 2 in embodiment 1 into each of the machining devices 4 and 5. In other words, each of the machining devices 4 and 5 has the functions of the processor 2a and storage device 2b in embodiment 1 as a standalone unit. The machining devices 4 and 5 are configured to acquire recommended values for the machining conditions taught by the teaching unit 16 of the processor 2a, and perform cutting using the acquired recommended values for the machining conditions.

The processing devices 4 and 5 applicable to this embodiment are various processing devices that perform cutting processes, such as lathes, machining centers, milling machines, gear cutting machines, and boring machines.

(others)
In the machining condition setting support system 100 of the second embodiment, all of the functions of the processor 2a and the storage device 2b of the first embodiment are incorporated into the machining devices 4 and 5, but some of the functions of the processor 2a and the storage device 2b may also be incorporated into the machining devices 4 and 5. In this case, the system may be configured to include a management device 2 and the machining devices 4 and 5 that form a network, with the management device 2 having some of the functions of the processor 2a and the storage device 2b, and the machining devices 4 and 5 incorporating the remaining functions.

1,100 Machining condition setting support system 2a Processor 2b Storage device 13 Stability limit diagram generation unit 14 Chatter occurrence determination unit Aa1 to Aa4 Cutting depth range SD1 to SD4 Stability limit diagram T Tool W Workpiece

Claims (9)

A machining condition setting support system including a processor, which evaluates or sets machining conditions including at least a cutting depth and a rotation speed when performing cutting while rotating at least one of a tool and a workpiece,
The processor:
a stability limit diagram generating unit that divides the depth of cut that can be taken as the machining condition into a plurality of sections, and that is capable of generating each stability limit diagram corresponding to each divided depth of cut section, the stability limit diagram defining the relationship between the rotation speed of the tool or the workpiece and the maximum depth of cut that does not cause chatter;
a chatter occurrence determination unit configured to determine whether chatter occurs in the case of an evaluation target cutting-in amount and an evaluation target rotation speed included in the machining conditions, using a target stability limit diagram corresponding to an evaluation target cutting-in amount among the plurality of stability limit diagrams that can be generated;
A processing condition setting support system equipped with the above. the processor and a storage device;
The storage device includes:
a compliance transfer function storage unit that stores a compliance transfer function that is a transfer function of a response displacement when an excitation force is applied to at least one of the tool and the workpiece,
The processor further comprises:
a specific cutting resistance output unit configured to output an estimated value of each specific cutting resistance, the specific cutting resistance being a value obtained by dividing the cutting resistance by the cutting cross-sectional area, the specific cutting resistance corresponding to each of the divided cutting depth sections, based on at least the cutting depth;
the stability limit diagram generating unit is configured to generate the stability limit diagram of an object corresponding to the cutting-in amount section including the cutting-in amount to be evaluated, based on the compliance transfer function and an estimated value of the specific cutting resistance of an object corresponding to the cutting-in amount to be evaluated as the machining condition,
2. The machining condition setting support system according to claim 1, wherein the chatter occurrence determination unit is configured to determine whether chatter occurs or not, using the stability limit diagram of the target generated by the stability limit diagram generation unit. The storage device further comprises:
a knowledge model storage unit for storing a knowledge model,
the knowledge model is configured to include a conditional probability table in which at least the cutting depth, the tool specifications, and the workpiece specifications among the machining conditions are parent factors, specific cutting resistance is a child factor, and a probability that the factor value of the child factor corresponds to a conditional factor value of the parent factor is defined;
3. The machining condition setting support system according to claim 2, wherein the specific cutting resistance output unit is configured to, when acquiring the factor value of the parent factor, output an estimated value of the specific cutting resistance as the child factor by performing forward probabilistic inference using the conditional probability table. In the conditional probability table, the cutting amount as the parent factor is discretized into each of the divided cutting amount intervals,
4. The machining condition setting support system according to claim 3, wherein each of the cutting depth ranges is set wide in a range where the change in the specific cutting resistance relative to the cutting depth is small, and is set narrow in a range where the change in the specific cutting resistance relative to the cutting depth is large. The cutting resistance generated in the cutting process is represented by a principal force, a thrust force, and a feed force,
4. The machining condition setting support system according to claim 3, wherein the specific cutting resistance corresponds to a value obtained by dividing the thrust force constituting the cutting resistance by a cutting cross-sectional area. the storage device further includes a prior distribution storage unit that stores parameters of a prior distribution that is preset to determine the conditional probability table;
The processor further comprises:
an observation data acquisition unit configured to acquire observation data obtained by performing the actual cutting process or a simulation of the cutting process, the observation data being about factor values of the parent factors and factor values of the child factors;
a conditional probability table generation unit configured to calculate a posterior distribution using Bayesian estimation based on the parameters of the prior distribution and the observation data, and to generate the conditional probability table in which each element value is an expected value of the posterior distribution;
Equipped with
4. The machining condition setting support system according to claim 3, wherein the knowledge model storage unit stores the knowledge model configured by the conditional probability table generated by the conditional probability table generation unit. the storage device further includes a weight storage unit configured to store weights representing influence ratios of parameters of the prior distribution to the observation data in the Bayesian estimation;
7. The machining condition setting support system according to claim 6, wherein the conditional probability table generation unit is configured to calculate the posterior distribution using the Bayesian estimation based on the observation data and parameters of the prior distribution taking into account the weights, and to generate the conditional probability table in which each element value is an expected value of the posterior distribution. The processor further comprises:
a recommended condition determination unit that determines recommended values of the machining conditions using the stability limit diagram;
an instruction unit that instructs the recommended values of the machining conditions determined by the recommended condition determination unit;
The processing condition setting support system according to any one of claims 1 to 7 , comprising: Further, a processing device configured to be able to communicate with the processor or configured with the processor incorporated therein is provided,
9. The machining condition setting support system according to claim 8 , wherein the machining device is configured to acquire the recommended values of the machining conditions taught by the teaching unit, and to perform the cutting processing by applying the acquired recommended values of the machining conditions.