JP7818824B2 - Manufacturing method of measuring device - Google Patents

Description

The present invention relates to a method for manufacturing a measuring device.

Near-infrared spectrometers are widely used as devices for sorting fruit and other agricultural products and measuring various parameters (such as sugar content). However, near-infrared spectrometers can have differences (instrumental error) between multiple devices, even if they are the same model, and even when measuring the same object using multiple devices, there can be variation in the measured values obtained. For this reason, it is necessary to ensure that nearly identical measured values are obtained between multiple devices. Furthermore, even with the same device, regular calibration is required to address changes in measured values over time.

When measuring agricultural products such as fruit, there are no standard samples whose characteristics are retained for long periods of time, so it is necessary to perform spectroscopic measurements on dozens of samples and analyze the target variables (such as sugar content and protein content) to confirm the estimated accuracy of the calibration curve. While calibration is relatively easy for items that are easy to analyze, such as sugar content, there are some quality items for which calibration is extremely difficult or impossible, such as calibration curves that estimate the taste and texture perceived by humans. This is a barrier to the development of agricultural product measurement using near-infrared spectrometers. In addition to near-infrared spectrometers, variations in sensor output characteristics can lead to variations in output characteristics between multiple measuring devices, which can make the measuring devices less reliable.

Patent No. 6080410 Japanese Patent Application Laid-Open No. 2021-139818

The present invention provides a method for manufacturing a measuring device that can reduce instrumental error between measuring devices.

To solve the above problem, the present invention provides a method for manufacturing a measuring device that includes multiple sensors and outputs an output signal based on the detection signals of the multiple sensors. This manufacturing method includes the steps of acquiring the output characteristics of each of N sensors, and selecting j sensors (N>>j) from the N sensors for one measuring device based on the output characteristics of each of the N sensors to configure multiple measuring devices.

The present invention provides a method for manufacturing a measuring device that can reduce instrumental error between measuring devices.

1 is a schematic diagram illustrating the configuration of a measuring device 10 manufactured by a manufacturing method according to a first embodiment. 10 is a graph illustrating the effect of providing a plurality of spectroscopic sensors 14 in one measurement device 10. 4 is a flowchart illustrating a method for manufacturing the measurement device according to the first embodiment. 1A to 1C are schematic diagrams illustrating an overview of a manufacturing method for a measurement device according to a first embodiment. 10 is a flowchart illustrating a method for manufacturing a measurement device according to a second embodiment. 10A and 10B are schematic diagrams illustrating an overview of a manufacturing method for a measuring device according to a second embodiment. 10 is a flowchart illustrating a method for manufacturing a measurement device according to a third embodiment. 10A and 10B are schematic diagrams illustrating an overview of a manufacturing method for a measurement device according to a third embodiment. 10 is a graph illustrating the effects of various embodiments. 10 is a graph illustrating the effects of various embodiments. 10 is a graph illustrating the effects of various embodiments. 10 is a flowchart illustrating a manufacturing method of a measurement device according to a modified example of the third embodiment. The results of applying principal component analysis to the detected spectral signals shown in FIG. 2 and then applying sensor selection by k-means clustering according to the third embodiment are shown below.

The present embodiment will be described below with reference to the accompanying drawings. In the accompanying drawings, functionally identical elements may be denoted by the same numerals. Note that the accompanying drawings show embodiments and implementation examples in accordance with the principles of the present disclosure, but these are intended to aid in understanding the present disclosure and should not be used to interpret the present disclosure in any limiting manner. The descriptions in this specification are merely typical examples and are not intended to limit the scope or application of the present disclosure in any way.

In this embodiment, the present disclosure has been described in sufficient detail to enable those skilled in the art to implement the present disclosure. However, it should be understood that other implementations and forms are possible, and that changes to the configuration and structure and substitutions of various elements are possible without departing from the scope and spirit of the technical ideas of the present disclosure. Therefore, the following description should not be interpreted as being limited to this.

[First embodiment]
First, a method for manufacturing a measuring device according to a first embodiment of the present invention will be described. Fig. 1 illustrates the configuration of a measuring device 10 manufactured by the manufacturing method according to the first embodiment. The upper view of Fig. 1 shows the structure of the measuring surface (lower surface) of the measuring device 10, and the lower view shows a side cross-sectional view.

This measuring device 10 is, as an example, a near-infrared spectroscopic device comprising a light source 11, an optical fiber 12, a light-shielding rubber 13, multiple spectroscopic sensors 14, and an entrance slit 15. The near-infrared spectroscopic device shown in Figure 1 is an example of a measuring device 10 that is the subject of the manufacturing method of the present invention. The light source 11 is a light source that emits measurement light to be irradiated onto the measurement object SP (fruit, etc.). The optical fiber 12 extends from the light source 11 toward the measurement surface (bottom surface) and is a light-guiding member that guides the measurement light toward the measurement surface.

The light-shielding rubber 13 is formed on the underside of the measuring device 10 so as to surround the output end face of the optical fiber 12, preventing the output light from entering the spectroscopic sensor 14 without passing through the measurement target SP (stray light).

The multiple spectroscopic sensors 14 are arranged, for example, at equal intervals on a circumference surrounding the optical fiber 12 on the measurement surface and measure the spectrum of the reflected light LB from the measurement target SP. Measuring the spectrum of the reflected light LB using the average or median of the detection values of the multiple spectroscopic sensors 14 reduces variations in characteristics among the multiple spectroscopic sensors 14 and thus performance variations among multiple measurement devices 10. The number of spectroscopic sensors 14 installed in one measurement device 10 is not limited to a specific number. Figure 1 illustrates an example in which one measurement device 10 is equipped with 16 spectroscopic sensors 14, each arranged at equal intervals of 22.5 degrees. The arrangement of the spectroscopic sensors 14 is also not limited to a specific arrangement. For example, instead of being arranged at equal intervals on the circumference as shown in Figure 1, the sensors can be arranged at equal intervals in a grid pattern, for example. An entrance slit 15 is provided in front of the multiple spectroscopic sensors 14 to limit the reflected light LB. The spectroscopic sensor 14 can be, for example, an ultra-compact grating spectrometer C14384MA-01 (product name) provided by Hamamatsu Photonics. Such spectrometers are capable of measuring light in the near-infrared range with high sensitivity, but there may be errors in sensitivity and in the correspondence between the installed detection element and the measured wavelength.

The detection signals from the multiple spectroscopic sensors 14 are input to a calculation control unit (not shown), and the average or median of the multiple detection values from the multiple spectroscopic sensors 14 can be output as the measurement value of the measuring device 10. By calculating the average or median of the detection values from the multiple spectroscopic sensors 14, even if there is variation in the performance of the spectroscopic sensors 14, it is possible to output a measurement value that eliminates the influence of that variation.

Figure 2 is a graph illustrating the effect of providing multiple spectroscopic sensors 14 in one measurement device 10. Figure 2(a) shows the variation in detected spectral signals between multiple measurement devices 10 when each measurement device 10 has a single spectroscopic sensor 14. Figure 2(b) shows the variation in detected spectral signals between multiple measurement devices 10 when each measurement device 10 has four spectroscopic sensors 14. Figure 2(c) shows the variation in detected spectral signals between multiple measurement devices 10 when each measurement device 10 has 16 spectroscopic sensors 14.

As shown in Figure 2, the more spectroscopic sensors 14 installed in one measuring device 10, the smaller the variation in detection signals between multiple measuring devices 10 becomes, since the average or median value of the detection values of the multiple spectroscopic sensors 14 is used as the output signal.

However, even if the number of spectroscopic sensors 14 mounted on one measuring device 10 increases, there will still be variation in the detection signals between measuring devices 10. Therefore, in this embodiment, we propose reducing this variation by adopting the manufacturing method described below.

The manufacturing method of the first embodiment will be described with reference to Figures 3 and 4. Figure 3 is a flowchart explaining the manufacturing method of the first embodiment, and Figure 4 shows an overview of the manufacturing method. In this method, to manufacture a measuring device 10 having j spectroscopic sensors 14 per unit, N spectroscopic sensors 14 (N>>j) are prepared, and these N spectroscopic sensors 14 are used as the population.

Then, the output characteristics of each of the multiple spectroscopic sensors 14 in this population are measured and acquired (step S11). The output characteristics may be measured by measuring the output signal of each spectroscopic sensor 14, or the output characteristics may be estimated based on visual inspection, etc. If the spectroscopic sensor 14 has multiple output channels, the output characteristics may be measured after performing dimensionality reduction using principal component analysis, t-distributed stochastic neighbor embedding (t-SNE), nonnegative matrix factorization (NMF), etc.

Suppose there are multiple (e.g., N) spectroscopic sensors 14, and information regarding their output characteristics is obtained in advance. N spectroscopic sensors 14 are then appropriately combined in groups of j (N>>j) to create multiple combinations of j spectroscopic sensors 14, and the output characteristics of multiple measuring devices 10 resulting from each combination are calculated on a computer. The output characteristics for all possible j combinations are then calculated "brute force." The combination that minimizes the variation in output characteristics among the multiple measuring devices 10 is then determined, and multiple measuring devices 10 each containing j spectroscopic sensors 14 are configured according to this combination (step S13).

According to the manufacturing method of this first embodiment, the output characteristics are measured in a round-robin manner for every possible combination of N spectroscopic sensors 14 that make up multiple measuring devices 10, each having j spectroscopic sensors 14, making it possible to uniform the output characteristics among the multiple measuring devices 10.

Second Embodiment
Next, a manufacturing method according to a second embodiment will be described with reference to FIGS. 5 and 6 . The manufacturing method according to the second embodiment also targets the manufacturing of a measuring device 10 as shown in FIG. 1 as an example. FIG. 5 is a flowchart illustrating the manufacturing method according to the second embodiment, and FIG. 6 shows an overview of the manufacturing method. This method also manufactures measuring devices 10 each having j spectroscopic sensors 14. To do this, N spectroscopic sensors 14 (N>>j) are prepared, and these N spectroscopic sensors 14 form a population. Then, the output characteristics of each of the N spectroscopic sensors 14 in this population are measured (step S21). Up to this point, the process is the same as in the first embodiment.

Then, in the manufacturing method of the second embodiment, the spectroscopic sensors 14 are grouped (classified) into multiple groups (e.g., k groups) according to their output characteristics. In this second embodiment, as an example, as shown in FIG. 6, groups (A) to (D) are defined in feature space, and grouping is performed so that spectroscopic sensors 14 with similar output characteristics are spaced apart (separated) in feature space (step S22). Then, j spectroscopic sensors 14 are selected from each of groups (A) to (D), and multiple measuring devices 10 each equipped with j spectroscopic sensors 14 are configured (step S23). Furthermore, the number of groups k is not particularly limited by j or N, and can be set to a value that allows optimal grouping.

As an example of an algorithm that can be used to select the spectroscopic sensor 14, the Kennard-Stone (KS) method and its derived algorithms can be used. This is an algorithm that first selects the two points that are furthest apart in the feature space, and then sequentially selects the point with the furthest minimum distance from all of the selected points. The KS method uses Euclidean distance to calculate distance, but derived algorithms may use Mahalanobis generalized distance.

In this second embodiment, spectroscopic sensors 14 with similar output characteristics are grouped so that they are spaced farther apart, and each of groups (A) to (D) can be configured as a collection of spectroscopic sensors 14 with different output characteristics. Therefore, by selecting spectroscopic sensors 14 that belong to the same group, the spectroscopic sensors 14 can be distributed among multiple measuring devices 10 without bias in output characteristics, making it possible to uniform the output characteristics among multiple measuring devices 10.

[Third embodiment]
Next, a manufacturing method according to a third embodiment will be described with reference to FIGS. 7 and 8 . The manufacturing method according to the third embodiment also targets the manufacturing of a measuring device 10 as shown in FIG. 1 as an example. FIG. 7 is a flowchart illustrating the manufacturing method according to the third embodiment, and FIG. 8 shows an outline of the manufacturing method. This method also manufactures measuring devices 10 each having j spectroscopic sensors 14. To do this, N spectroscopic sensors 14 (N>>j) are prepared, and these N spectroscopic sensors 14 form a population. Then, the output characteristics of each of the N spectroscopic sensors 14 in this population are measured (step S31). Up to this point, the process is the same as in the first and second embodiments.

Then, in the manufacturing method of the third embodiment, the spectroscopic sensors 14 are grouped (classified) into multiple groups according to the output characteristics of the spectroscopic sensors 14. As an example, similar to the second embodiment, the grouping can be performed by defining four groups (A) to (D) in four quadrants of the feature space.

The difference from the second embodiment is that, whereas in the second embodiment, grouping is performed so that spectroscopic sensors 14 with similar output characteristics are farther apart in feature space, in this third embodiment, grouping is performed so that spectroscopic sensors with similar output characteristics are in the same group (step S32). Then, a predetermined number of spectroscopic sensors 14 are selected from each of the multiple groups (A) to (D) to configure multiple measurement devices 10, each having j spectroscopic sensors 14 (step S33). It is preferable that the number of spectroscopic sensors 14 selected to configure one measurement device 10 from each of the multiple groups (A) to (D) be the same across groups, but depending on the conditions, the number selected can be different from one group to another.

In this third embodiment, spectroscopic sensors 14 with similar output characteristics are grouped together, so that each of groups (A) to (D) can be configured as a collection of spectroscopic sensors 14 with similar output characteristics. Therefore, by selecting a predetermined number of spectroscopic sensors from each of groups (A) to (D), the spectroscopic sensors 14 can be distributed among multiple measuring devices 10 without bias in output characteristics, making it possible to uniform the output characteristics among multiple measuring devices 10.

Algorithms such as hierarchical clustering, k-means clustering, DBSCAN (Density-Based Spatial Clustering of Application with Noise), k-nearest neighbor search, and spectral clustering can be used to classify the spectroscopic sensors 14 into groups.

Figure 9 is a graph illustrating the effects of the second and third embodiments. Specifically, the upper graph in Figure 9 shows the variation in output characteristics among multiple measuring devices 10 when measuring devices 10 each have j spectroscopic sensors 14 assembled by randomly combining N spectroscopic sensors 14 without grouping them. In the upper graph, small dots indicate the output characteristics (output values 1 and 2) of each spectroscopic sensor 14, and large round dots indicate the output characteristics of the measuring device 10. When random combinations are performed, as shown in the upper graph, the output characteristics of some measuring devices 10 may differ significantly compared to other measuring devices 10.

The bottom graph in Figure 9 shows the variation in output characteristics among multiple measurement devices 10 when N spectroscopic sensors 14 are grouped and then the spectroscopic sensors 14 are selected using the Kennard-Stone method. It can be seen that the variation in output characteristics among multiple measurement devices 10 is suppressed compared to the top graph (when the devices are randomly combined).

The three graphs in Figure 10 show the variation in output characteristics among multiple measurement devices 10 when N spectroscopic sensors 14 are grouped and then the spectroscopic sensors 14 are selected using hierarchical cluster analysis (top graph), K-means cluster analysis (middle graph), and spectral clustering (bottom graph). It can be seen that the variation in output characteristics among multiple measurement devices 10 is suppressed compared to the top graph in Figure 9 (when the devices are randomly combined).

Figure 11 is a graph comparing the standard deviations of the output values (output value 1, output value 2) of the measurement device 10. As shown in Figure 11, the standard deviation is large when the spectroscopic sensors 14 are randomly combined, but the standard deviation is smaller when the spectroscopic sensors 14 are selected from each group using the method of this embodiment (Knard-Stone method, hierarchical cluster analysis, K-means cluster analysis, and spectral clustering).

In the third embodiment, after grouping of N spectroscopic sensors has been performed, a new spectroscopic sensor other than the N spectroscopic sensors may be added to the population. In this case, it is not necessary to perform new grouping of all N spectroscopic sensors belonging to the population; grouping can be performed only on the spectroscopic sensor to be newly added. Figure 12 is a flowchart illustrating a manufacturing method of a modified example of the third embodiment when a new spectroscopic sensor is added to the population.

First, the output characteristics of each of the spectroscopic sensors 14 newly added to the population are measured (step S34). Next, for groups consisting of N existing (already grouped) spectroscopic sensors, for example, four groups (A) to (D), the average output characteristics of the spectroscopic sensors belonging to each group are calculated (step S35).

Then, each newly added spectroscopic sensor 14 is placed in the group whose output characteristics are closest to the average output characteristics of the group (step S36). Finally, a predetermined number of spectroscopic sensors 14 are selected from each of the multiple groups (A) to (D) to configure multiple measurement devices 10, each with j spectroscopic sensors 14 (step S37).

Note that, although the second and third embodiments have been described with examples using two output values, the number of output values is not limited. Furthermore, if there are a large number of output values and it is difficult to grasp the output characteristics of each sensor, the number of dimensions of the output characteristics may be compressed in advance using techniques such as principal component analysis, t-SNE (t-distribution stochastic neighbor embedding), or NMF (non-negative matrix factorization).

Figure 13 shows the results of applying principal component analysis to the detected spectral signals shown in Figure 2, followed by sensor selection using k-means clustering according to the third embodiment. The original detected spectral signal is composed of 1100 nm to 2498 nm at 2 nm intervals, with 700 output values. However, principal component analysis compresses the signal into two-dimensional principal components that represent the characteristics of each spectroscopic sensor, making it possible to visually grasp the differences in output characteristics between spectroscopic sensors.

As shown in Figure 13, it can be seen that the variation in output characteristics among multiple measuring devices 10 is smaller when sensor selection using k-means clustering is applied compared to when N spectroscopic sensors 14 are not grouped and instead N spectroscopic sensors 14 are randomly combined to assemble a measuring device 10 having j spectroscopic sensors 14 per device.

(others)
In the embodiment described above, the measuring device 10 including a plurality of spectroscopic sensors 14 as sensors has been described as an example. However, the spectroscopic sensors 14 have been described as an example, and the present invention can also be applied to other types of sensors in general, where there is variation in output characteristics among the plurality of sensors. For example, by applying a method similar to that described above to a method for manufacturing a measuring device having a plurality of thermocouples as sensors, it is possible to reduce variation in output characteristics among the plurality of measuring devices.

The present invention is not limited to the above-described examples, and various modifications are possible. For example, the above-described examples have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to embodiments that include all of the described configurations. Furthermore, it is possible to replace part of the configuration of one example with the configuration of another example. It is also possible to add the configuration of another example to the configuration of one example. It is also possible to delete part of the configuration of each example, or to add or replace other configurations.

REFERENCE SIGNS LIST 10...measuring device, 11...light source, 12...optical fiber, 13...light-shielding rubber, 14...spectroscopy sensor, 15...entrance slit, SP...measurement object, LB...measurement light

Claims (7)

A method for manufacturing a measuring device that includes a plurality of sensors and outputs an output signal based on detection signals of the plurality of sensors, comprising:
acquiring output characteristics of each of the N sensors;
and selecting j sensors (N>>j) for one measurement device from among the N sensors based on the output characteristics of each of the N sensors to configure a plurality of measurement devices ;
The step of selecting j sensors includes:
calculating output characteristics of all combinations of selecting the j sensors from the N sensors;
selecting a combination that minimizes the variation in the output characteristics of the plurality of measuring devices;
A method for manufacturing a measuring device. A method for manufacturing a measuring device that includes a plurality of sensors and outputs an output signal based on detection signals of the plurality of sensors, comprising:
acquiring output characteristics of each of the N sensors;
and selecting j sensors (N>>j) for one measurement device from among the N sensors based on the output characteristics of each of the N sensors to configure a plurality of measurement devices ;
further comprising a step of classifying the plurality of sensors into a plurality of groups according to output characteristics of the plurality of sensors such that sensors having similar output characteristics are spaced apart from each other in a feature space;
the step of selecting j sensors includes selecting j sensors in each of the plurality of groups.
A method for manufacturing a measuring device. A method for manufacturing a measuring device that includes a plurality of sensors and outputs an output signal based on detection signals of the plurality of sensors, comprising:
acquiring output characteristics of each of the N sensors;
and selecting j sensors (N>>j) for one measurement device from among the N sensors based on the output characteristics of each of the N sensors to configure a plurality of measurement devices ;
further comprising a step of classifying the plurality of sensors into a plurality of groups according to output characteristics of the plurality of sensors such that sensors having similar output characteristics are grouped together, and the step of selecting the j number of sensors includes selecting a predetermined number of sensors from each of the plurality of groups.
A method for manufacturing a measuring device. The manufacturing method according to claim 2 , wherein the step of selecting the j sensors for the one measurement device to configure the plurality of measurement devices uses the Kennard-Stone (KS) method and its derived algorithms. 4. The method for manufacturing a measurement device according to claim 3 , wherein the step of classifying the plurality of sensors into a plurality of groups includes hierarchical clustering, k-means clustering, DBSCAN (Density-Based Spatial Clustering of Application with Noise), k-nearest neighbor search, or spectral clustering. After the N sensors are classified into a plurality of groups, when a new spectroscopic sensor is added to the population of the N spectroscopic sensors,
The step of classifying the plurality of sensors into a plurality of groups includes:
measuring the output characteristics of each newly added sensor to the population;
calculating an average output characteristic of the sensors belonging to the plurality of groups;
4. The method of claim 3 , further comprising the step of classifying the newly added sensor into one of the plurality of groups based on a relationship between its output characteristic and the average output characteristic of the plurality of groups. 4. The method for manufacturing a measurement device according to claim 1, wherein the step of acquiring the output characteristics of each of the N sensors includes a step of applying principal component analysis, t-SNE (t-distribution type stochastic neighbor embedding), or NMF (non-negative matrix factorization) to the output characteristics of the N sensors to reduce the dimensionality of the output characteristics.