JPH0327067B2 - - Google Patents

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention is designed to strike a tile 8 attached to a wall of a building, receive the striking sound generated at the time, and use the received signal to generate a striking sound. The present invention relates to a method for detecting peeling of wall tiles, which determines the state of peeling of tiles.

[Prior Art] It is dangerous when tiles, etc. attached to the wall of a building with mortar peel off and fall. Therefore, the present applicant developed a conventional peeling detection device that detects the state of adhesion of tiles, etc. by the sound of impact. By special public service Showa 56-20498
Publication No. 57-23221, Publication No. 53-
This is proposed in Publication No. 1992, etc.

Although these known techniques are effective in themselves, (a) small peels as small as 5 cm in diameter or 7 cm deep;
It is difficult to detect peeling that is deep within a cm of the surface; (b) It takes time to move the detection device, and it is difficult to perform this on walls with many uneven surfaces; (c)
The drawback was that it required time and labor to process data and create drawings.

Therefore, in Japanese Patent Application No. 59-67470, the present applicant proposed a peeling detection device that can detect the size and depth of peeling, is easy to move, and can easily visualize data.

In order to better understand the present invention, the apparatus proposed in the aforementioned Japanese Patent Application No. 59-67470 will first be explained with reference to FIGS. 1 and 1a.

Figure 1 shows the tile T attached to the wall W of building B.
The part where the degree of peeling is to be detected is shown, and the suspended moving part, which is indicated by reference numeral 1, is fixed by engaging with both sides of the parapet P formed on the roof of building B. A plurality of fixing fittings 2,
A guide rail 3 fixed on the plurality of fixing fittings 2 and extending above the parapet P, and wheels 4 and engaging portions 5 that run on the guide rail 3.
and a hanging unit 6 formed in a U-shape as a whole. A pair of rods 7, 7 extending downward are fixed to the wall W side of the hanging unit 6, and guide wheels 8, which engage with the side surfaces of the parapet P, respectively, are attached to the rods 7, 7.
8 is rotatably provided, and a pair of arms 9, 9 extending from the lower side of the parapet P toward the wall surface W are provided at the lower ends thereof, respectively, so as to be rotatable in a horizontal plane. Pulleys 10, 10 are provided at the ends of the arms 9, 9 on the wall surface W side, respectively.
A guide wire 1 is attached to these pulleys 10, 10.
1 and 11 are wound respectively.

These guide wires 11, 11 are attached with a striking part, the entirety of which is indicated by the reference numeral 12.
A drive unit 13 is provided at the lower ends of the wires 11, 11. Therefore, by operating a motor (not shown) provided in the drive unit 13, the guide wire 11 moves, and the movement of the guide wire 11 causes an impact. The portion 12 is adapted to move up and down. Since the striking part 12 moves up and down along the wall surface W as described above, it has an appropriate number of guide rollers 14.

As shown in FIG. 1a, the striking section 12 is provided with a striking unit U. This striking unit U
1, a guide rod 15 extending laterally from approximately the center of the striking part 12, a moving part 16 that moves in the left-right direction in the drawing along the guide rod 15 by a motor (not shown), and a striking part provided on the moving part 16. It is provided with a hammer 17, a load detector 18 provided at the tip of the striking hammer 17, and a microphone 19 for detecting the striking sound when the tile T is struck with the striking hammer 17.

Therefore, the vertical movement of the striking part 12 and the moving part 1
Due to the left and right movement of 6, the striking hammer 17 hits the wall surface W.
can be inspected in a planar manner. The above drive unit 1
3 and the movement command of the moving part 16 and the load detector 18
and the signal from the microphone 19 is transmitted to the control unit 2.
0 is sent and received. In other words, the control unit 20 is the control panel 2
1, and various operations are performed on an operation panel 22.
The control unit 20 includes a monitor 23, a floppy disk 24, and a printer 2 for data processing.
5 and a host computer 26.

The hanging unit 6, which is movably installed on the parapet P of the building B, is horizontally movable as described above, but to make the explanation easier to understand, the inspection range of the wall W is shown as a pair of wires 11, 11, and therefore the suspension unit 6 is assumed to be fixed.

The striking part 12 is moved to a predetermined position by a movement command signal from the control part 20, and the movable body 16 is moved to a predetermined position. Then, the striking position detector 30 detects the striking positions X and Y of the striking hammer 17 and sends the coordinate signals to the host computer 26. On the other hand, due to the impact operation of the impact hammer 17, the load detector 18 sends the load, that is, the impact force, and the microphone 19 sends the impact sound to the peeling determining section 31. This peeling determination section determines the size and depth of the peeled portion, and the determination signal is sent to the host computer 26. In this way, the determination results for each position (each coordinate) between the wires 11, 11 are stored in the host computer 26, and the results are stored in the floppy disk 24, displayed on the monitor 23, and printed by the printer 25. be done.

On the other hand, with the recent development of electronic devices, voice discrimination devices are being used in various fields. An example of this is shown in FIG. 2. In the preprocessing circuit 31, the sound from the microphone 30 is amplified and filtered to increase the level of a specific frequency.
An AD conversion circuit 32 performs AD conversion, a processing circuit 33 collects frequencies from the frequency characteristics of the voice, generates parameters, compresses the parameters as necessary, and registers this as a pattern in a storage circuit 34. In this way, the memory circuit 34
Various sounds are registered as patterns. Next, when another voice is similarly input to the determination circuit 35 as an input pattern, the comparison circuit 36 compares the pattern registered in the storage circuit 34 with the determination circuit 35.
If the pattern distance is relatively small, the input voice is determined to match the registered pattern, and if there is a pattern distance, the next pattern is determined (step S1). Next, the presence or absence of the next pattern is checked (step S2), and if there is, it is compared again, and if it is not present or if it matches the pattern in step S1, the result is displayed on the display device (step S3). In this way, the input voice is displayed.

Now, if the voice recognition port shown in Fig. 2 is used in the tile peeling detection device shown in Fig. 1,
The comparison work is relatively simple and reliable.

However, in the case of voice, for example, the normal pronunciation of "a" lasts about 0.3 seconds, but in the case of a percussion sound, it lasts only about 10 milliseconds. Therefore, if the impact sound is discriminated by a normal commercially available voice recognition means, it will be inaccurate and the correct state of peeling, that is, the position, size, etc. of the peeling cannot be determined.

[Object of the Invention] Therefore, an object of the present invention is to provide a method for detecting peeling of a wall tile, which can accurately determine the state of peeling of a tile from the impact sound using ordinary voice recognition means.

[Structure of the Invention] According to the present invention, a received impact sound signal is temporarily stored, the temporarily stored impact waveform signal is time-divided into a plurality of sections, and the waveform signal within each section is inverted every time. This is repeated as many times as necessary to generate a time-expanded signal, which is then further subjected to Walsh-Hadamard transformation to obtain the auto power spectrum for each sequence band and compared with the registered signal. .

[Operation and Effect of the Invention] Therefore, by folding and repeating the waveform of the characteristic part, the time is extended while connecting smoothly, and furthermore, this signal is subjected to Wall-Hadamard transform to obtain an auto power spectrum for the sequence band, and this Since comparison and discrimination are performed based on patterns, quick and accurate discrimination can be achieved.

[Embodiments] Hereinafter, embodiments of the present invention will be described with reference to the drawings from FIG. 4 onwards.

FIG. 4 shows an example of a circuit implementing the present invention, in which an input signal X1 from a striking hammer is input to a circuit 51, amplified by an amplifier 60, and taken out as a signal V. This signal V and a predetermined reference signal _V0 are stored in the memory 61, and when the signal V is larger than _V0 , the peak hold circuit 62 is activated to hold the peak value of the signal X1, and AD It is processed by a converter 63 and sent to a parallel input/output port 64 which acts as an interface. At the same time, the input signal X2 from the microphone is pre-processed through the amplifier 65, high-pass filter 66, and band-limiting filter 67 of the circuit 52, held in the sample-and-hold circuit 68, and digitized through the AD converter 69, which acts as an interface. The data is sent to a parallel input/output port 70 and input to a random access memory (RAM) 71. In addition to the RAM 71, the control device includes a ROM 72, a central control unit (CPU) 73, and a bus 74, and is configured to transmit an output 56 via a parallel input/output port 75.

FIG. 5 shows an example of the input waveform X2, in which the horizontal axis shows time and the vertical axis shows amplitude. First, when implementing the present invention. This waveform is divided into appropriate time segments from the beginning. This example shows a case where the time is divided into four time segments A, B, C, and D every 6.4 milliseconds. Next, this divided waveform is subjected to time expansion processing, and this processing is performed as follows. now,
If the original waveform generated in section A is W1/W2, after this waveform W1/W2, waveform W1/W2 is generated.
Waveforms W2/W1 are generated by folding back W2 and connected. Original waveform W follows waveform W2/W1
Connect 1/W2 and repeat this process. This waveform inversion process allows the amplitudes of continuous parts to be smoothly connected. FIG. 6 shows a case where eight original waveforms and folded waveforms are connected, and the time period of the original section A of 6.4 milliseconds is expanded to 51.2 milliseconds. The same processing is performed for sections B, C, and D, and in the whole of FIG. 6, a signal having a wavelength time sufficient for inputting 204.8 milliseconds, or approximately 0.2 seconds, to the speech recognition board is generated.

Next, this signal is subjected to the Walsh-Hadamard transform method to obtain the auto power spectrum (APS) shown in FIG. In Figure 7, the X-axis represents time, the Y-axis represents sequence, or frequency, and the Z-axis represents time.
Its strength is shown on the axis. This auto power spectrum is divided into bands according to the frequency of the waveform within each section, for example from 200 Hz to 5000 Hz.
Divide the hertz frequency into sequential bands of 1/3 octave to obtain the spectrum for each band.

This auto power spectrum is defined as section A,
FIG. 7 shows four frames representing each of B, C, and D. That is, each frame forms a matrix of sequence bands and spectra. On the other hand, process the signal obtained by hitting a normal tile wall surface to obtain a similar matrix pattern, register it, and compare this registered pattern with the pattern obtained in Fig. 7. It is possible to detect peeling of tile walls. Furthermore, by focusing on a specific octave section of each frame, a spectrum matrix for this section can be obtained, and this matrix becomes data indicating the characteristics of the shock response wave. Therefore, by comparing the spectrum matrices for each octave section, it is possible to more accurately detect a normal tile wall and a tile peeling part.

Next, the manner in which waveform X2 is processed will be specifically explained with reference to FIG.

First, start the RAM of the primary storage device 45.
The waveform X2 input to 71 has sections A, B, C, and D.
Each section is stored as data, and each section of the data is divided into i sections. First, the number of pieces to be taken out is set to zero (step S10), and then the central control section 73 attaches the first (i+1) piece (step S10).
S11), determine whether or not to return (step
S12). In the case of the first i pieces, there is no need to loop back (NO), so the data is output and sent to the voice discrimination circuit 47 (step S13). If you need to turn around, change direction (step S14) and step S13.
send to Next, the number N input from the RAM 74 is counted (step S15), and it is determined whether all data have been output (step S16).
This number N is i×4 (A, B, C, D).
If NO, it is determined whether all i data of section A have been output (step S17), and if NO, the next data (i+1) is extracted. If YES, all the data of section A has been output, so the next data of section B is input, and the process returns to step S10. In this way, all sections A,
If the data of B, C, and D are output, step
YES in S16 and the work ends.

[Summary] As described above, the present invention divides the first characteristic part of the response waveform obtained by hitting a tile wall surface into time sections, and extends the time by folding back and connecting the waveforms within these sections. Furthermore, this time-stretched waveform is subjected to the Wallace-Hadamard transform to obtain an auto power spectrum, and this auto power spectrum pattern is used to compare with the pattern of a normal tile wall surface, making for a more accurate comparison. Moreover, since commercially available voice recognition equipment can be used, it is possible to provide an inexpensive, small, and lightweight detection system.

[Brief explanation of the drawing]

Fig. 1 is an explanatory diagram of a wall tile peeling detection device which is the basis of the present invention, Fig. 1a is an explanatory diagram of its striking part, Fig. 2 is a block diagram showing a known voice discrimination device, and Fig. 3 is a diagram of the present invention. FIG. 4 is a block diagram showing details of the main parts of the block diagram shown in FIG. 3; FIG. 5 shows a waveform obtained by time-stretching the impact sound waveform. 7 is a diagram showing a power spectrum obtained by performing Walsh-Hadamard transform on the waveform of FIG. 6, and FIG. 8 is a diagram showing a flowchart. 42...Microphone, 45...Temporary memory circuit, 46
. . . Repetitive waveform generation circuit, 47 . . . Sound discrimination circuit, 51 . . . Hit load input processing circuit, 52 .
Central processing unit.

Claims (1)

[Claims] 1. In a wall tile peeling detection method in which a tile installed on a wall of a building is hit, the impact sound generated at that time is received, and the state of tile peeling is determined based on the received signal. By temporarily storing the signal of the impact sound caused by the impact, dividing the temporarily stored impact waveform signal into multiple sections, inverting and connecting the waveform signals in each section, and repeating this as many times as necessary. generate a time-stretched signal,
A method for detecting peeling of a wall tile, which is characterized in that the time-stretched signal is further subjected to Wallace-Hadamard transformation to obtain an auto power spectrum for each sequence band and compared with a registered signal.