WO2021149614A1 - Wire rope flaw detection device - Google Patents

Abstract

In this wire rope flaw detection device, a magnetizer produces magnetic flux that passes through a portion of a wire rope. A magnetic sensor generates, as a sensor signal, a signal corresponding to leakage magnetic flux that is the magnetic flux that has leaked from the wire rope. A filter unit extracts frequency components of the sensor signal. A computation unit extracts a plurality of feature values based on the plurality of values of the frequency components. A learning unit determines whether there is a flaw in a wire included in the wire rope through computation processing by a learning model that has carried out machine learning of the correlation between the plurality of feature values and the state of the wires included in the wire rope and has had the plurality of feature values input therein.

Description

Wire rope flaw detector

The present invention relates to a wire rope flaw detector.

Conventionally, a wire rope flaw detector having a magnetizer that magnetically saturates the wire rope and a magnetic sensor that detects a leakage magnetic flux leaking from the wire rope due to a damaged portion of the wire rope is known (for example, Patent Document). 1).

Japanese Unexamined Patent Publication No. 09-210966

The amount of leakage magnetic flux leaking from the wire rope decreases as the wire rope becomes thinner. Therefore, in the conventional wire rope flaw detector shown in Patent Document 1, the thinner the wire rope, the smaller the amount of leakage magnetic flux reaching the magnetic sensor, and the lower the output of the magnetic sensor. As a result, the signal-to-noise ratio of the conventional wire rope flaw detector is reduced.

Further, in the conventional wire rope flaw detector, even if the SN ratio is improved by shortening the distance between the wire rope and the magnetic sensor, there are restrictions such as the assembly accuracy of the wire rope and the magnetic sensor. Therefore, the conventional wire rope flaw detector has a limit in reducing the distance between the wire rope and the magnetic sensor.

The present invention has been made to solve the above problems, and an object of the present invention is to obtain a wire rope flaw detector capable of more reliably improving the SN ratio.

The wire rope flaw detector according to the present invention includes a magnetizer that generates a magnetic flux that passes through a part of the wire rope, and a magnetic sensor that generates a signal corresponding to the leakage magnetic flux leaking from the wire rope as a sensor signal. The control unit includes a control unit that processes the sensor signal, and the control unit extracts a filter unit that extracts a frequency component of the sensor signal and a plurality of feature quantities based on a plurality of values constituting the frequency component. A plurality of frequency components extracted by the filter unit are configured in a trained learning model in which the calculation unit and the trained learning model in which the correlation between the plurality of feature quantities and the state of the wire rope included in the wire rope is machine-learned. The learning model when a plurality of feature quantities extracted by the calculation unit are input based on the value of the above has a learning unit for determining the presence or absence of damage to the wire rope by executing the calculation process. There is.

According to the wire rope flaw detector according to the present invention, the SN ratio can be improved more reliably.

It is an exploded perspective view which shows the probe of the wire rope flaw detector according to Embodiment 1. FIG. It is explanatory drawing which shows the flaw detection principle by the probe of FIG. It is an enlarged view of the part A of FIG. It is a figure explaining more concretely an example of the positional relationship between the leakage magnetic flux and a coil of FIG. It is a figure explaining more concretely an example of the positional relationship between a leakage magnetic flux leaking from a wire rope whose diameter is smaller than that of the wire rope of FIG. 4 and a coil. It is a block diagram which shows the functional structure example of the control part of FIG. It is a figure which shows an example of the frequency characteristic of the filter part of FIG. It is a figure which shows an example of the distribution of the frequency component extracted from the input signal by the filter part of FIG. It is a figure which shows an example of the sequence yk (n) generated by the filter part at the time t1 of FIG. It is a figure which shows an example of the sequence yk (n) generated by the filter part at the time t2 of FIG. It is a figure which shows the concept of performing machine learning using a learning data set when the learning part of FIG. 6 is configured as a support vector machine. 6 is a flowchart illustrating processing by the control unit when the learning unit of FIG. 6 is configured as a support vector machine. FIG. 6 is a diagram showing a concept of performing machine learning using a learning data set when the learning unit of FIG. 6 is configured as an autoencoder. 6 is a flowchart illustrating processing by the control unit when the learning unit of FIG. 6 is configured as an autoencoder. It is a flowchart explaining the process by a control part in Embodiment 2. FIG. 5 is a block diagram showing a functional configuration example of a control unit that processes a signal corresponding to a leakage magnetic flux leaking from a wire rope according to a third embodiment. It is a figure which shows an example of the frequency characteristic of the filter part of FIG. It is a figure which shows the waveform example of the time domain of the mother wavelet by the wavelet transform part of FIG. It is a figure which shows the waveform example of the frequency domain of the mother wavelet by the wavelet transform part of FIG. As another example of the frequency characteristics of the filter unit of FIG. 16, it is a conceptual diagram of the center frequency at the time of 1/3 octave. It is a figure which shows an example of the distribution of the frequency component extracted from the input signal by the filter part of FIG. It is a figure which shows an example of the sequence yk (n) generated by the filter part at the time t1 of FIG. It is a figure which shows an example of the sequence yk (n) generated by the filter part at the time t2 of FIG. It is a flowchart explaining the process by the control part of FIG. It is a figure explaining the hardware configuration example. It is a figure explaining another hardware configuration example. 25 is a diagram showing a system configuration example in which at least one of the control units of FIGS. 6 and 16 is incorporated into a terminal device as a specific example of FIG. 25 is a diagram showing a system configuration example in which at least one of the control units of FIGS. 6 and 16 is incorporated into the determination device 401 as a specific example of FIG. 25 or FIG. 26 to supply the processing contents of the determination device to the data logger. 25 is a diagram showing a system configuration example in which at least one of the control units of FIGS. 6 and 16 is incorporated into the determination panel as a specific example of FIG. 25 or FIG.

Embodiment 1.
FIG. 1 is an exploded perspective view showing a probe 1 of the wire rope flaw detector according to the first embodiment. The probe 1 includes a probe main body 3 and a cover 5.

The cover 5 is made of a non-magnetic material. The cover 5 covers the probe body 3. As a result, the cover 5 protects the probe body 3. The cover 5 is provided with a groove 51. The cross section of the groove 51 is formed in a U shape. The groove portion 51 has a first end portion 51_1 and a second end portion 51_2.

The probe main body 3 includes a magnetizer 11 and a magnetic sensor 13.

The magnetizer 11 has a back yoke 111, a first permanent magnet 112_1, a second permanent magnet 112_2, a first pole piece 113_1, and a second pole piece 113_2.

The back yoke 111 is made of a ferromagnetic material. The back yoke 111 has a first yoke end portion 111_1, a second yoke end portion 111_2, and a yoke central portion 111_3. One end of the back yoke 111 in the longitudinal direction is a first yoke end 111_1. The other end of the back yoke 111 in the longitudinal direction is a second yoke end 111_2. The yoke central portion 111_3 is located between the first yoke end portion 111_1 and the second yoke end portion 111_2.

A first pole piece 113_1 is fixed to the first yoke end 111_1 via a first permanent magnet 112_1. A second pole piece 113_2 is fixed to the second yoke end 111_2 via a second permanent magnet 112_2. As a result, the first permanent magnet 112_1 and the second permanent magnet 112_2 are arranged apart from each other in the longitudinal direction of the back yoke 111. Further, the first pole piece 113_1 and the second pole piece 113_1 are arranged apart from each other in the longitudinal direction of the back yoke 111.

The first pole piece 113_1 is made of a ferromagnetic material. The first pole piece 113_1 is provided with a first pole piece groove 113_1. The cross section of the first pole piece groove portion 113_11 is formed in a U shape. The first pole piece groove portion 113_1 is fixed to the cover 5 at a position on the back side of the first end portion 51_1.

The second pole piece 113_2 is made of a ferromagnetic material. The second pole piece 113_2 is provided with a second pole piece groove 113_21. The cross section of the second pole piece groove portion 113_21 is formed in a U shape. The second pole piece groove portion 113_21 is fixed to the cover 5 at a position on the back side of the second end portion 51_2.

The first permanent magnet 112_1 is arranged between the first pole piece 113_1 and the first yoke end portion 111_1. The first permanent magnet 112_1 has one magnetic pole surface oriented toward the first pole piece 113_1 and the other magnetic pole surface directed toward the first yoke end 111_1. As the first permanent magnet 112_1, for example, a neodymium magnet is used. The first permanent magnet 112_1 generates a magnetomotive force.

The second permanent magnet 112_2 is arranged between the second pole piece 113_2 and the second yoke end 111_2. The second permanent magnet 112_2 is arranged with one magnetic pole surface facing the second yoke end 111_2 and the other magnetic pole surface facing the second pole piece 113_2. As the second permanent magnet 112_2, for example, a neodymium magnet is used. The second permanent magnet 112_2 generates a magnetomotive force.

The magnetic sensor 13 has a sensor body 13A and a mounting portion 13B.

The mounting portion 13B is mounted on the yoke central portion 111_3. The mounting portion 13B is made of a non-magnetic material.

The sensor body 13A is arranged between the first pole piece 113_1 and the second pole piece 113_2. The sensor body 13A has a base portion 132, a coil holder 133, a first coil 131_1, and a second coil 131_2.

The base portion 132 is attached to the attachment portion 13B. The coil holder 133 is attached to the base portion 132. The coil holder 133 is made of a ferromagnetic material. The first coil 131_1 and the second coil 131_2 are attached to the coil holder 133.

FIG. 2 is an explanatory diagram showing the flaw detection principle by the probe 1 of FIG. The wire rope flaw detector includes a probe 1 and a control unit 9 that receives a signal from the probe 1.

In FIG. 2, for convenience of illustration, the outline of the cover 5 is shown by a chain double-dashed line. Further, in FIG. 2, for convenience of illustration, the cross-sectional shape portion of the groove portion 51 is shown by hatching. When the wire rope flaw detection device performs a flaw detection inspection of the wire rope 2, the wire rope 2 moves with respect to the probe 1 in a specific direction W_D along the longitudinal direction of the groove 51. The probe 1 performs measurement while bringing the wire rope 2 into contact with the groove 51.

In the example of FIG. 2, the polar direction of the first permanent magnet 112_1 is the direction from the first yoke end 111_1 to the first pole piece 113_1. Further, in one example of FIG. 2, the polarity direction of the second permanent magnet 112_2 is the direction from the second pole piece 113_2 toward the second yoke end portion 111_2.

That is, the polarity of the first permanent magnet 112_1 is opposite to the polarity of the second permanent magnet 112_2. Therefore, in the state where the wire rope 2 is arranged in the groove 51, the magnetic flux F passing through the magnetic circuit F_C composed of a part of the wire rope 2 and the magnetizer 11 is transferred to the first permanent magnet 112_1 and the second permanent magnet. 112_2 occurs.

As a result, when the wire rope 2 is arranged in the groove 51, the section W between the portion of the wire rope 2 facing the first pole piece 113_1 and the portion facing the second pole piece 113_2 The wire rope 2 is magnetized at. In the wire rope 2, the magnetic flux F generated by the first permanent magnet 112_1 and the second permanent magnet 112_2 passes along the longitudinal direction of the wire rope 2. That is, the magnetizer 11 generates a magnetic flux F that passes through a part of the wire rope 2.

The magnetic sensor 13 generates a signal corresponding to the leakage magnetic flux L_F leaking from the wire rope 2 among the magnetic flux F as a sensor signal. The control unit 9 processes the sensor signal generated from the magnetic sensor 13. A detailed description of the magnetic flux F and the leakage magnetic flux L_F will be described later with reference to FIGS. 3, 4, and 5.

Hereinafter, the first permanent magnet 112_1 and the second permanent magnet 112_2 are collectively referred to as the permanent magnet 112. Further, the first pole piece 113_1 and the second pole piece 113_2 are collectively referred to as a pole piece 113. Further, the first coil 131_1 and the second coil 131_2 are collectively referred to as the coil 131.

Next, the principle of detecting the leakage magnetic flux L_F by the wire rope flaw detector will be described with reference to FIGS. 3, 4, and 5. FIG. 3 is an enlarged view of part A of FIG. As shown in FIG. 3, if there is a damaged portion B_W in the portion of the wire rope 2 through which the magnetic flux F passes, a part of the magnetic flux F leaks from the wire rope 2 as a leakage magnetic flux L_F around the damaged portion B_W. ..

FIG. 4 is a diagram for more specifically explaining an example of the positional relationship between the leakage magnetic flux L_F of FIG. 3 and the coil 131. When the wire rope 2 is moved with respect to the probe 1, the first coil 131_1 and the second coil 131_2 are interlinked with the leakage magnetic flux L_F. Therefore, an induced voltage, which is a signal corresponding to the leakage magnetic flux L_F, is generated in the first coil 131_1 and the second coil 131_2 as sensor signals.

By the way, the wire rope 2 is composed of a core rope and a plurality of strands 21 twisted around the core rope at a constant pitch λ. Therefore, on the outer peripheral portion of the wire rope 2, a plurality of convex portions arranged in the length direction of the wire rope 2 are formed at a constant pitch λ. Further, the strand 21 is formed by twisting a plurality of strands into a single layer or multiple layers. Therefore, if the wire contained in the wire rope 2 is thin, the diameter of the wire rope 2 is reduced.

FIG. 5 is a diagram for more specifically explaining an example of the positional relationship between the leakage magnetic flux L_F leaking from the wire rope 2S having a diameter smaller than that of the wire rope 2 of FIG. 4 and the coil 131. Leakage interlinking with the first coil 131_1 and the second coil 131_2 when the wire rope 2S of FIG. 5 moves with respect to the probe 1 than when the wire rope 2 of FIG. 4 moves with respect to the probe 1. The amount of magnetic flux L_F is reduced. Therefore, the control unit 9 in FIG. 2 distinguishes whether the sensor signal generated from the magnetic sensor 13 is caused by the magnetic flux F due to noise or the magnetic flux F due to the damaged part B_W. Difficult to put on. Therefore, in the present embodiment, the control unit 9 of FIG. 2 is a machine for the correlation between a plurality of feature quantities based on a plurality of values constituting the frequency component of the sensor signal and the state of the wire rope included in the wire rope. After learning, it is judged whether or not the wire is damaged.

As described above, the wire rope 2 is composed of strands 21 twisted at a constant pitch λ. Therefore, the probe 1 detects noise caused by the outer peripheral portion of the wire rope 2 at least every pitch λ. Further, the wire rope 2S is also configured by being twisted at a pitch λS in the same manner. Therefore, a plurality of convex portions arranged in the length direction of the wire rope 2S are similarly formed on the outer peripheral portion of the wire rope 2S at a constant pitch λS. Therefore, the probe 1 detects noise caused by the outer peripheral portion of the wire rope 2S at least for each pitch λS.

FIG. 6 is a block diagram showing a functional configuration example of the control unit 9 of FIG. As shown in FIG. 6, the control unit 9 includes a measuring instrument 91, a synthesizer 92, a filter unit 93, and a processing unit 94.

The measuring instrument 91 has a first measuring instrument 91_1 and a second measuring instrument 91_2. The first measuring instrument 91_1 is connected to both ends of the first coil 131_1. The second measuring instrument 91_2 is connected to both ends of the second coil 131_2. In this example, the first coil 131_1 is located upstream of the second coil 131_2 in the specific direction W_D of the wire rope 2S.

As shown in FIG. 6, when the damaged portion B_W enters between the first pole piece 113_1 and the second pole piece 113_1, the leakage magnetic flux L_F leaks from the wire rope 2 around the damaged portion B_W. The leakage magnetic flux L_F is interlinked with the first coil 131_1 and then with the second coil 131_2. Therefore, the time at which the peak of the induced voltage generated at both ends of the first coil 131_1 occurs is shifted by the delay time τ from the time at which the peak of the induced voltage generated at both ends of the second coil 131_1 occurs. The delay time τ is represented by a value obtained by dividing the distance P between the centers of the first coil 131_1 and the second coil 131_2 by the moving speed ν of the probe 1.

Therefore, the first measuring instrument 91_1 detects the induced voltage, which is a signal corresponding to the leakage magnetic flux L_F leaking from the wire rope 2S, as the sensor signal f1 (t−τ). Further, the second measuring instrument 91_2 detects the induced voltage, which is a signal corresponding to the leakage magnetic flux L_F leaking from the wire rope 2S, as the sensor signal f2 (t).

The synthesizer 92 superimposes the sensor signal f1 (t−τ) detected by the first measuring instrument 91_1 and the sensor signal f2 (t) detected by the second measuring instrument 91_1 to form the sensor signal x. (T) is generated. Specifically, the synthesizer 92 detects the sensor signal f1 (t−τ) detected by the first measuring instrument 91_1 with the sensor signal f1 (t) delayed by the time τ and the second measuring instrument 91_2. The generated sensor signal f2 (t) is superimposed. As a result, the sensor signal x (t) becomes a signal obtained by combining the peaks of the induced voltage generated across the first coil 131_1 and the peaks of the induced voltage generated across the second coil 131_2.

The synthesizer 92 samples the sensor signal x (t) at a constant period Ts. Since the sampled signal is a signal for each period Ts, the time can be represented by an integer n with the period Ts as a unit. That is, the relationship between the analog time of the sensor signal x (t) and the time of the sampled signal is t = n · Ts. The amplitude of the sampled signal is real. Therefore, the discrete signal, which is a sampled signal, is represented as a sequence of real values {x (0), x (1), x (2), ...}, So it is represented as a sequence x (n). To. The sequence x (n) is supplied to the filter unit 93 as an input signal x (n).

Hereinafter, the sequence x (n) is referred to as an input signal x (n) of the filter unit 93.

The filter unit 93 extracts the frequency component of the input signal x (n) that samples the sensor signal x (t). The filter unit 93 has a plurality of FIR (Finite Impulse Response) filters 931 as a plurality of bandpass filters, and a plurality of absolute value units 932.

FIG. 7 is a diagram showing an example of the frequency characteristics of the filter unit 93 of FIG. As shown in FIG. 7, the number of taps, the gain, and the bandwidth b are constant in each of the plurality of FIR filters 931. The plurality of FIR filters 931 have a plurality of bands different from each other as individual pass bands. That is, the plurality of FIR filters 931 have different pass bands different from each other. Therefore, the filter unit 93 extracts the frequency component of the input signal x (n) in each of the plurality of bands different from each other.

As shown in FIG. 6, the plurality of absolute value units 932 obtain the absolute values of the frequency components of the input signals x (n) supplied from the plurality of FIR filters 931. Here, the input signal x (n) is a sequence of real values {x (0), x (1), x (2), ...} As described above. Therefore, the plurality of absolute value units 932 are quantized by dividing the input signal x (n) by a quantization unit and rounding off to obtain a sequence yk (n) of integer values. However, k is a value in ascending order from 1 to N. Also, N is a natural number.

For example, when a real value x (0) is input to the plurality of FIR filters 931, the plurality of FIR filters 931 extract the frequency component of the real value x (0) in each of a plurality of bands different from each other. The absolute value unit 932 performs the above calculation on the frequency component of the real value x (0) for each individual pass band, so that the integer values y1 (0), y2 (0), ..., And yN (0) Ask for.

Hereinafter, the integer values y1 (0), y2 (0), ... And yN (0) are referred to as a sequence of integer values {y1 (0), y2 (0), ... And yN (0)}. It is expressed as a sequence yk (0).

The filter unit 93 performs the same processing on the real value x (1) to obtain yk (1). The filter unit 93 performs the same processing on the real value x (2) and later, and obtains yk (2) and later. From the above description, the filter unit 93 obtains the sequence yk (n) from the input signal x (n).

FIG. 8 is a diagram showing an example of the distribution of frequency components extracted from the input signal x (n) by the filter unit 93 of FIG. As shown in FIG. 8, the moving speed ν of the wire rope 2S is trapezoidally controlled, for example. Therefore, when the wire rope 2S is moving at a constant velocity, the periodic fluctuation of the input signal x (n) due to the shape of the outer peripheral portion of the wire rope 2S is constant. Periodic variation of the input signal x (n) due to the shape on the outer peripheral portion of the wire rope 2S occurs for each pitch λS of the strand 21S as described above. Therefore, the constant periodic fluctuation of the input signal x (n) occurs at a specific frequency.

For example, as shown in FIG. 8, the noise frequency component f_n of the input signal x (n) appears in the bands when k = 6 and k = 8 when the wire rope 2S is moving at a constant velocity. do.

Therefore, the frequency component of the constant periodic fluctuation of the input signal x (n) can be regarded as the noise frequency component f_n of the input signal x (n) among the frequency components of the input signal x (n).

Further, when the input signal x (n) is a signal synthesized by the synthesizer 92 according to the leakage magnetic flux L_F, the signal generated during the local minute time Δt and the input signal x (n) in the time domain. ) Is equivalent. Therefore, when the input signal x (n) is a signal synthesized by the synthesizer 92 according to the leakage magnetic flux L_F, the shorter the minute time Δt, the more the frequency component of the input signal x (n) appears in the band. The number increases.

For example, as shown in FIG. 8, the damage frequency component f_s of the input signal x (n) is k = 3, k = 4, k = 5, k = 6, k = 7, k = 8 and k = 9. Appears in each band of time.

As a result, the distribution of the frequency components of the input signal x (n) becomes a distribution over a plurality of bands. Therefore, among the frequency components of the input signal x (n), the damaged frequency component f_s of the input signal x (n) also appears in a band other than the band in which the noise frequency component f_n of the input signal x (n) appears.

By the way, the damaged frequency component f_s of the input signal x (n) originally appears even when k = 1 and k = 2. However, here, it is not necessary to consider until k = 1 and k = 2. Therefore, in one example of FIG. 8, when k = 1 and k = 2, the damaged frequency component f_s of the input signal x (n) is cut off in the process of various signal processing.

The distribution of frequency components is composed of a sequence yk (n). As described above, the sequence yk (n) is composed of integer values y1 (n), y2 (n), ..., And yN (n). Therefore, the distribution of frequency components is composed of a plurality of values y1 (n) to yN (n).

Further, as shown in the middle side and the lower side of FIG. 8, each sequence yk (n) when k = 6 and k = 7 can be arranged over time. On the middle side of FIG. 8, an example is shown in which the control unit 9 obtains the crest factor from the sequence yk (n) when k = 6. The crest factor is a value obtained by calculating the ratio of the maximum value to the effective value. Further, on the lower side of FIG. 8, an example is shown in which the control unit 9 obtains the derivative or the difference from the sequence yk (n) when k = 7.

As shown in FIG. 6, the processing unit 94 has a calculation unit 941 and a learning unit 943.

The calculation unit 941 extracts a plurality of feature quantities based on a plurality of values y1 (n) to yN (n). Specifically, the calculation unit 941 extracts a plurality of feature quantities from a plurality of values y1 (n) to yN (n) by a plurality of statistical calculations. The plurality of feature quantities are representative values that characterize the frequency components of the input signal x (n). The statistical operation is, for example, an operation for finding the median value. The median value is a value located at the center when the integer values y1 (n), y2 (n), ..., And yN (n) constituting the sequence yk (n) are arranged in ascending order. In addition, statistical calculations include, for example, maximum value, minimum value, range, average value, standard deviation, effective value, crest factor, differentiation, and difference calculation. Further, among the statistical operations, the range is a value obtained by an operation for obtaining the difference between the maximum value and the minimum value.

FIG. 9 is a diagram showing an example of the sequence yk (n) generated by the filter unit 93 at the time t1 of FIG. In an example of FIG. 9, among a plurality of values y1 (n) to yN (n) constituting the sequence yk (n), the value y8 (n) when k = 8 indicates the maximum value, and k = 9. The value y9 (n) at the time indicates the minimum value. Therefore, the range can be obtained by the difference between the value y8 (n) and the value y9 (n).

FIG. 10 is a diagram showing an example of the sequence yk (n) generated by the filter unit 93 at the time t2 of FIG. In one example of FIG. 10, when a plurality of values y1 (n) to yN (n) constituting the sequence yk (n) are arranged in ascending order, the value located at the center is the value when k = 5. Further, in FIG. 10, in addition to the median value, the average value, the standard deviation, and the derivative or the difference are shown.

As shown in FIG. 6, the learning unit 943 inputs a plurality of feature quantities based on a plurality of values y1 (n) to yN (n) into the learning model 961. The learning model 961 when a plurality of feature quantities are input executes arithmetic processing. The learning unit 943 determines whether or not the wire rope included in the wire rope 2S is damaged from the execution result of the arithmetic processing of the learning model 961. In the learning model 961, machine learning is performed on the correlation between a plurality of features based on a plurality of values y1 (n) to yN (n) constituting the frequency component and the state of the wire contained in the wire rope 2S. It has already been learned.

The learning unit 943 inputs a plurality of feature quantities into a learning model 961 in which at least one of the plurality of first feature quantities and the plurality of second feature quantities is used as a learning data set, and executes arithmetic processing. From the output of the model 961, it is determined whether or not the wire rope 2S is damaged. Here, the plurality of first feature quantities are a set of a plurality of values that characterize the presence of damage to the wire contained in the wire rope 2S. Further, the plurality of second feature quantities are a set of a plurality of values that characterize that there is no damage to the wire contained in the wire rope 2S.

Further, the damage to the wire contained in the wire rope 2S is the physical damage caused to at least a part of the wire rope 2S. Physical damage is, for example, damage to at least one of wire breakage, partial wire breakage, and scratch marks on the wire.

Hereinafter, damage to the wire contained in the wire rope 2S will be appropriately referred to as damage to the wire.

<Support vector machine>
FIG. 11 is a diagram showing a concept of performing machine learning using a learning data set when the learning unit 943 of FIG. 6 is configured as a support vector machine 943_1. The support vector machine 943_1 sets a plurality of first feature quantities and a plurality of second feature quantities as inputs as an example of machine learning of supervised learning.

Specifically, as shown in FIG. 11, the learning unit 943 of FIG. 6 can classify the preset feature space into a first region and a second region by an identification hyperplane. It is configured as machine 943_1.

Here, the first region is a region including a plurality of first feature quantities. Specifically, the first region is a region including the first support vector SV_1 which is a part of the plurality of first feature quantities. The second region is a region including a plurality of second feature quantities. Specifically, the second region is a region including the second support vector SV_2, which is a part of the plurality of second feature quantities. The discriminating hyperplane is, for example, a ^{function that can be represented by w T} Γ + b = 0, as shown in FIG.

More specifically, the plurality of first feature quantity is represented by the sequence gamma _'q. q is 1, 2, ..., a N _{e. Ne} is the number of samples with damage to the wire. Series gamma _'q are multiple values γ'1 (n), γ'2 (n ), and a... And γ'M (n).

For example, γ'1 (n) is assigned the maximum value in the sequence yk (n). Further, the minimum value in the sequence yk (n) is assigned to γ'2 (n). Thus, the sequence gamma _'q is constructed from the values computed value is assigned by a plurality of statistical operation.

Therefore, the identification in the hyperplane capable of becoming characterized space there is w ^T Γ + b = 0, when a portion of the plurality of first feature quantity and the first support vector SV_1, series gamma _'q is the first support vector SV_1 as an expression containing, for example, represented by _{^{w T Γ 'q + b =}} 1.

In other words, support vector machine 943_1 represents 'what the _q mapped onto the feature space w ^T gamma' sequence gamma function of _q + b = 1.

On the other hand, the plurality of second features are represented by the _{sequence Γ p.} p is 1, 2, ..., _No. N _o is the number of samples although the damage of the wire is not. The sequence Γ _p is composed of a plurality of values γ1 (n), γ2 (n), ..., And γM (n).

For example, γ1 (n) is assigned the maximum value in the sequence yk (n). Further, the minimum value in the sequence yk (n) is assigned to γ2 (n). In this way, the sequence Γ _p is composed of values to which the values calculated by a plurality of statistical operations are assigned.

^{Therefore, on a feature space with w T} Γ + b = 0, which can be an identification hyperplane _{, the sequence Γ p} includes the second support vector SV_2, where a part of the plurality of second features is the second support vector SV_2. As an equation, for example, ^{it is expressed by w T} Γ _p + b = -1.

That is, the support vector machine 943_1 represents a mapping of the sequence Γ _p to the above feature space by a ^{function of w T} Γ _p + b = -1.

Incidentally, support vector machine 943_1 is within a value constituting the series gamma _'q, sets the closest to one of the values that constitute the sequence gamma _p to the first support vector SV_1. Also, support vector machine 943_1 is within a value constituting the sequence gamma _p, sets the closest to one of the values that constitute the series gamma _'q in the second support vector SV_2.

Further, the support vector machine 943_1 ^{obtains w T} and b having a distance d1 and a distance d2. Here, the distance d1 is the distance at which w ^T Γ + b = 0 is the farthest from the first support vector SV_1. On the other hand, the distance d2 is the distance at which w ^T Γ + b = 0 is the farthest from the second support vector SV_2. The support vector machine 943_1 calculates the identification hyperplane by finding ^{such w T and b.}

That is, the support vector machine 943_1 generates a learning model 961_1 by mapping both a plurality of first feature quantities and a plurality of second feature quantities to the feature space as a training data set and calculating an identification hyperplane.

Specifically, the support vector machine 943_1 considers that the wire is damaged when the output of the learning model 961-1 when a plurality of features are input to the learning model 961-1 corresponds to the one classified in the first region. judge. Further, the support vector machine 943_1 determines that there is no damage to the strands when the output of the learning model 961_1 when a plurality of feature quantities are input to the learning model 961_1 corresponds to those classified into the second region.

FIG. 12 is a flowchart illustrating processing by the control unit 9 when the learning unit 943 of FIG. 6 is configured as the support vector machine 943_1. The processes of steps S11 to S13 are learning processes. The processes of steps S14 and S15 are feature extraction processes. The processes of steps S16 to S20 are strand determination processes. The process of step S21 and step S22 is a period determination process.

<Learning process>
In step S11, the support vector machine 943_1 determines whether or not the training data set has been input. When the support vector machine 943_1 determines that the training data set has been input, the support vector machine 943_1 shifts the current processing to the processing in step S12. When the support vector machine 943_1 determines that the training data set has not been input, the process of step S11 is continued.

In step S12, the support vector machine 943_1 determines whether or not both the plurality of first feature quantities and the plurality of second feature quantities are included in the training data set. When the support vector machine 943_1 determines that the training data set includes both the plurality of first feature quantities and the plurality of second feature quantities, the current process shifts to the process of step S13. When the support vector machine 943_1 determines that the training data set does not include both the plurality of first feature quantities and the plurality of second feature quantities, the current process is returned to the process of step S11.

In step S13, the support vector machine 943_1 calculates a discriminant hyperplane that can be classified into a first region and a second region by performing machine learning using the learning data set, and generates a learning model 961_1. The support vector machine 943_1 shifts the current process to the process of step S14.

<Feature extraction process>
In step S14, the calculation unit 941 determines whether or not the frequency component of the sensor signal x (t) has been extracted. When the calculation unit 941 determines that the frequency component of the sensor signal x (t) has been extracted, the calculation unit 941 shifts the current process to the process of step S15. When the calculation unit 941 determines that the frequency component of the sensor signal x (t) has not been extracted, the calculation unit 941 repeats the process of step S14. That is, the process of step S14 is a process in which the calculation unit 941 determines whether or not the frequency component of the sensor signal x (t) has been extracted by the filter unit 93.

In step S15, the calculation unit 941 extracts a plurality of feature quantities from a plurality of values y1 (n) to yN (n) constituting the frequency component by a plurality of statistical calculations. The calculation unit 941 shifts the current processing to the processing in step S16.

<Strand wire judgment processing>
In step S16, the support vector machine 943_1 inputs a plurality of feature quantities into the learning model 961-1. The support vector machine 943_1 shifts the current process to the process of step S17. In step S17, the support vector machine 943_1 determines whether or not the output of the learning model 961_1 corresponds to a plurality of feature quantities classified into the first region. When the support vector machine 943_1 determines that the output of the learning model 961_1 corresponds to the one in which a plurality of feature quantities are classified into the first region, the support vector machine 943_1 shifts the current processing to the processing in step S18. When the support vector machine 943_1 determines that the output of the learning model 961_1 does not correspond to the one in which a plurality of feature quantities are classified into the first region, the support vector machine 943_1 moves the current processing to the processing in step S19.

In step S18, the support vector machine 943_1 determines that the wire is damaged. The support vector machine 943_1 shifts the current process to the process of step S21. In step S19, the support vector machine 943_1 determines whether or not the output of the learning model 961_1 corresponds to a plurality of feature quantities classified into the second region. When the support vector machine 943_1 determines that the output of the learning model 961_1 corresponds to a plurality of feature quantities classified into the second region, the support vector machine 943_1 shifts the current processing to the processing in step S20. When the support vector machine 943_1 determines that the output of the learning model 961_1 does not correspond to the one in which a plurality of feature quantities are classified into the second region, the support vector machine 943_1 shifts the current processing to the processing in step S21.

In step S20, the support vector machine 943_1 determines that the wire is not damaged. The support vector machine 943_1 shifts the current process to the process of step S21.

<Period judgment processing>
In step S21, the filter unit 93 determines whether or not to end the determination of the presence or absence of damage to the wire. When the filter unit 93 determines that the determination of the presence or absence of damage to the wire is completed, the filter unit 93 ends the current process. When the filter unit 93 determines that the determination of the presence or absence of damage to the wire is not completed, the filter unit 93 shifts the current process to the process of step S22. The filter unit 93 determines whether or not the speed of the wire rope 2S corresponds to the constant velocity moving period. When the filter unit 93 determines that the speed of the wire rope 2S corresponds to the constant velocity movement period, the filter unit 93 returns the current process to the process of step S14. When the filter unit 93 determines that the speed of the wire rope 2S does not correspond to the constant velocity movement period, the filter unit 93 continues the process of step S22.

<Autoencoder>
FIG. 13 is a diagram showing a concept of performing machine learning using a learning data set when the learning unit 943 of FIG. 6 is configured as an autoencoder 943_2.

Specifically, as shown in FIG. 13, the learning unit 943 of FIG. 6 is configured as an autoencoder 943_2 that suppresses the error between the input and the output to a constant amount by the weighting coefficient wf, the weighting coefficient wg, and the weighting coefficient wh. There is. The autoencoder 943_2 sets a plurality of second feature quantities as inputs as an example of machine learning by unsupervised learning.

More specifically, the autoencoder 943_2 has a hierarchical structure formed by laminating a part of the autoencoder 962_1, a part of the autoencoder 962_2, and the final layer 965.

First, the autoencoder 962_1 has an encoder 963_1 and a decoder 964_1. For the autoencoder 962_1, for example, a sigmoid function is used as the activation function. The autoencoder 943_2 sets the input of the encoder 963_1 to the input of the autoencoder 962_1. The autoencoder 943_2 sets the output of the encoder 963_1 to the input of the decoder 964_1. The autoencoder 943_2 sets the output of the decoder 964_1 to the output of the autoencoder 962_1. The autoencoder 943_2 calculates the weighting coefficient wf, the weighting coefficient wg, and the weighting coefficient wh by learning so that the same one input to the autoencoder 962_1 is reconstructed on the output side of the autoencoder 962_1.

Next, the autoencoder 943_2 sets the output of the encoder 963_1, which is a part of the autoencoder 962_1, to the input of the autoencoder 962_2. The autoencoder 962_2 has an encoder 963_2 and a decoder 964_2. For the autoencoder 962_2, for example, a sigmoid function is used as the activation function. The autoencoder 943_2 sets the input of the encoder 963_2 to the input of the autoencoder 962_2. The autoencoder 943_2 sets the output of the encoder 963_2 to the input of the decoder 964_2. The autoencoder 943_2 sets the output of the decoder 964_2 to the output of the autoencoder 962_2. Similar to the autoencoder 962_2, the autoencoder 943_2 learns the weighting coefficient wf, the weighting coefficient wg, and the weighting coefficient wh so that the same input to the autoencoder 962_2 is reproduced on the output side of the autoencoder 962_2. Calculate.

Next, the autoencoder 943_2 sets the output of the encoder 963_2, which is a part of the autoencoder 962_2, to the input of the final layer 965. Therefore, the autoencoder 943_2 sets the output of the encoder 963_1 to the input of the encoder 963_2. The autoencoder 943_2 adds a final layer 965 to the output side of the encoder 963_2. For the final layer 965, the softmax function is used as the activation function. Next, the autoencoder 943_2 sets the learning model 961_2 by finely adjusting the weighting coefficient wf, the weighting coefficient wg, and the weighting coefficient wh in the hierarchical structure of the encoder 963_1, the encoder 963_2, and the final layer 965 by using the error back propagation method. Generate.

That is, the autoencoder 943_2 learns by calculating the weighting coefficient wf, the weighting coefficient wg, and the weighting coefficient wh by using a plurality of second feature quantities for each of the input and the output of the autoencoder 943_2 as a learning data set. Generate model 961-2.

Further, the autoencoder 943_2 determines that there is no damage to the strands when the output of the learning model 961_2 when a plurality of feature quantities are input to the learning model 961_2 can reconstruct the plurality of feature quantities. The autoencoder 943_2 determines that the wire is damaged when the output of the learning model 961_2 when a plurality of feature quantities are input to the learning model 961_2 cannot reconstruct the plurality of feature quantities.

Specifically, the autoencoder 943_2 sets the difference between the plurality of feature quantities and the output of the learning model 961_2 when the plurality of feature quantities are input to the learning model 961_2 as the reconstruction error. The autoencoder 943_2 calculates the root mean square error by square averaging the reconstruction error. When the root mean square error exceeds the error tolerance, the autoencoder 943_2 determines that the output of the learning model 961_2 could not reconstruct a plurality of features. When the root mean square error is equal to or less than the error tolerance, the autoencoder 943_2 determines that the output of the learning model 961_2 has been able to reconstruct a plurality of features. The error tolerance is a value set according to a plurality of feature quantities.

FIG. 14 is a flowchart illustrating processing by the control unit 9 when the learning unit 943 of FIG. 6 is configured as an autoencoder 943_2. The processing of steps S41 to S43 is a learning process. The processes of steps S44 and S45 are feature extraction processes. The processes of steps S46 to S49 are strand determination processes. The processing of step S50 and step S51 is a period determination processing. Of the learning process, feature amount extraction process, wire line determination process, and period determination process, the period determination process is the same as the processes of steps S21 and S22 of FIG. Therefore, the description thereof will be omitted.

<Learning process>
In step S41, the autoencoder 943_2 determines whether or not the training data set has been input. When the autoencoder 943_2 determines that the training data set has been input, the autoencoder 943_2 shifts the current process to the process of step S42. When the autoencoder 943_2 determines that the learning data set has not been input, the autoencoder 943_2 continues the process of step S41.

In step S42, the autoencoder 943_2 determines whether or not the learning data set contains a plurality of second feature quantities. When the autoencoder 943_2 determines that the learning data set contains a plurality of second feature quantities, the autoencoder 943_2 shifts the current process to the process of step S43. When the autoencoder 943_2 determines that the training data set does not include the plurality of second feature quantities, the autoencoder 943_2 returns the current processing to the processing in step S41.

In step S43, the autoencoder 943_2 calculates the weighting coefficient wf, the weighting coefficient wg, and the weighting coefficient wh that suppress the error between the input and the output to a certain amount by performing machine learning using the learning data set, and is a learning model. Generate 961-2. The autoencoder 943_2 shifts the current process to the process of step S44.

<Feature extraction process>
In step S44, the calculation unit 941 determines whether or not the frequency component of the sensor signal x (t) has been extracted. When the calculation unit 941 determines that the frequency component of the sensor signal x (t) has been extracted, the calculation unit 941 shifts the current process to the process of step S45. When the calculation unit 941 determines that the frequency component of the sensor signal x (t) has not been extracted, the calculation unit 941 continues the process of step S44.

In step S45, the calculation unit 941 extracts a plurality of feature quantities from a plurality of values y1 (n) to yN (n) constituting the frequency component by a plurality of statistical calculations. The calculation unit 941 shifts the current processing to the processing in step S46.

<Strand wire judgment processing>
In step S46, the autoencoder 943_2 inputs a plurality of feature quantities into the learning model 961_2. The autoencoder 943_2 shifts the current process to the process of step S47. In step S47, the autoencoder 943_2 determines whether or not the output of the learning model 961_2 has been able to reconstruct a plurality of feature quantities. When the autoencoder 943_2 determines that the output of the learning model 961_2 has been able to reconstruct a plurality of feature quantities, the autoencoder shifts the current process to the process of step S48. When the autoencoder 943_2 determines that the output of the learning model 961_2 has not been able to reconstruct a plurality of feature quantities, the autoencoder shifts the current process to the process of step S49.

In step S48, the autoencoder 943_2 determines that there is no damage to the wire. The autoencoder 943_2 shifts the current process to the process of step S50.

In step S49, the autoencoder 943_2 determines that the wire is damaged. The autoencoder 943_2 shifts the current process to the process of step S50.

From the above description, the wire rope flaw detector includes a magnetizer 11, a magnetic sensor 13, and a control unit 9. The magnetizer 11 generates a magnetic flux F that passes through a part of the wire rope 2S. The magnetic sensor 13 generates a signal corresponding to the leakage magnetic flux L_F leaking from the wire rope 2S in the magnetic flux F as a sensor signal x (t). The control unit 9 processes the sensor signal x (t).

The control unit 9 has a filter unit 93 for extracting the frequency component of the sensor signal x (t), a calculation unit 941, and a learning unit 943.

The learning unit 943 generates a trained learning model 961 as a preliminary step for determining whether or not there is damage to the wire. The trained learning model 961 performs machine learning on the correlation between a plurality of features based on a plurality of values y1 (n) to yN (n) constituting the frequency component and the state of the wire rope included in the wire rope 2S. Is going.

The learning unit 943 determines whether or not there is damage to the strands by executing arithmetic processing in the learning model 961 when a plurality of feature quantities are input to the learned learning model 961. Here, the plurality of feature quantities are extracted based on a plurality of values y1 (n) to yN (n) constituting the frequency components extracted at a timing different from the previous step of determining the presence or absence of damage to the strands. There is.

That is, the processing unit 94 performs arithmetic processing of the learning model 961 when the input of a plurality of feature quantities based on the plurality of values y1 (n) to yN (n) constituting the frequency component is used as the learned learning model 961. The presence or absence of damage to the wire is determined from the execution result. Here, the learning model 961 is machine learning about the correlation between a plurality of features based on a plurality of values y1 (n) to yN (n) constituting the frequency component and the state of the wire rope included in the wire rope 2S. Have been learned.

Therefore, the learning model 961 can input a plurality of features based on a plurality of values y1 (n) to yN (n) constituting the frequency component to the wire rope 2S without performing a complicated calculation. It is possible to output the prediction result of predicting the state.

Therefore, in the time domain, even if the SN ratio is low due to the low induced voltage generated in the magnetic sensor 13, the result is as follows. The processing unit 94 can determine the presence or absence of damage to the strands in a state where the noise frequency component f_n is concealed by the output from the learning model 961 input based on the frequency component in the frequency domain.

From the above explanation, the wire rope flaw detector can more reliably improve the SN ratio.

Further, the calculation unit 941 extracts a plurality of feature quantities from a plurality of values y1 (n) to yN (n) by a plurality of statistical calculations. The learning unit 943 uses at least one of a plurality of first feature quantities that characterize the presence of damage to the strands and a plurality of second feature quantities that characterize the absence of damage to the strands as a learning data set. From the output of the learning model 961, it is determined whether or not the wire is damaged.

That is, the learning unit 943 is a learning model 961 when a plurality of feature quantities are input to the learning model 961 in which at least one of the plurality of first feature quantities and the plurality of second feature quantities is used as a learning data set. Based on the output of, it is judged whether or not the wire is damaged. Here, the plurality of first feature quantities characterize that there is damage to the wire. The plurality of second feature quantities are characterized by no damage to the strands.

Therefore, the learning unit 943 can clearly determine the presence or absence of damage to the strands from a side surface different from the leakage magnetic flux L_F in the time domain. Therefore, the wire rope flaw detector can more reliably improve the signal-to-noise ratio.

Further, the learning unit 943 can classify the preset feature space into a first region including a plurality of first feature quantities and a second region including a plurality of second feature quantities by an identification hyperplane. It is configured as a support vector machine 943_1. The support vector machine 943_1 generates a learning model 961_1 by mapping both a plurality of first feature quantities and a plurality of second feature quantities to a feature space as a training data set and calculating an identification hyperplane.

Therefore, the support vector machine 943_1 can classify by the learning model 961_1 on the identification hyperplane whether a plurality of features correspond to the case where the wire is damaged or the case where the wire is not damaged. Therefore, the wire rope flaw detector can improve the generalization ability for determining the presence or absence of damage to the wire.

Further, the support vector machine 943_1 determines that the wire is damaged when the output of the learning model 961_1 when a plurality of feature quantities are input to the learning model 961_1 corresponds to the one classified in the first region. The support vector machine 943_1 determines that there is no damage to the strands when the output of the learning model 961_1 when a plurality of feature quantities are input to the learning model 961_1 corresponds to those classified into the second region.

Therefore, the support vector machine 943_1 determines the presence or absence of wire damage according to whether the output of the learning model 961_1 corresponds to a plurality of feature quantities classified into either the first region or the second region. do. Therefore, the support vector machine 943_1 can determine whether or not the wire is damaged depending on which region it belongs to. Therefore, the wire rope flaw detector can replace the determination process of presence / absence of damage to the wire with a simple classification process, so that erroneous determination can be reduced.

Further, the learning unit 943 is configured as an autoencoder 943_2 that suppresses the error between the input and the output to a certain amount by the weighting coefficient wf, the weighting coefficient wg, and the weighting coefficient wh. The autoencoder 943_2 uses a plurality of second features as a training data set for each of the input and output of the autoencoder 943_2 to calculate the weighting coefficient wf, the weighting coefficient wg, and the weighting coefficient wh, thereby calculating the training model 961_2. To generate.

Therefore, the autoencoder 943_2 can indicate by output whether a plurality of features correspond to the case where the wire is damaged or the case where the wire is not damaged. Therefore, the autoencoder 943_2 can reduce the number of dimensions by weighting each of the plurality of feature quantities and determine the presence or absence of damage to the strands.

From the above description, the wire rope flaw detector can reduce the amount of calculation required for determination, and emphasizes the feature amount that contributes to the determination of the presence or absence of damage to the wire among a plurality of feature amounts. It is possible to improve the accuracy of determining the presence or absence of damage.

Further, the autoencoder 943_2 determines that there is no damage to the strands when the output of the learning model 961_2 when a plurality of feature quantities are input to the learning model 961_2 can reconstruct the plurality of feature quantities. The autoencoder 943_2 determines that the wire is damaged when the output of the learning model 961_2 when a plurality of feature quantities are input to the learning model 961_2 cannot reconstruct the plurality of feature quantities.

Therefore, the autoencoder 943_2 determines whether or not the wire is damaged depending on whether or not the output of the learning model 961_2 can reconstruct a plurality of feature quantities. Therefore, the wire rope flaw detector can determine the presence or absence of damage to the wire even if each of the plurality of features has a non-linear relationship.

Note that the autoencoder 943_2 may execute the following processing when it is determined by the learning model 961_2 that the wire is damaged. The autoencoder 943_2 estimates by sparse optimization which input dimension is responsible for determining that the wire is damaged. As a result, the autoencoder 943_2 can use the estimation result for a more detailed analysis of the broken portion of the wire rope 2.

Further, the filter unit 93 has a plurality of bandpass filters having a plurality of bands different from each other as individual pass bands. Therefore, the filter unit 93 can extract frequency components of a plurality of bands different from each other. Therefore, the wire rope flaw detector can analyze the induced voltage generated by the magnetic sensor 13 as the sensor signal x (t) in the frequency domain.

Further, the calculation unit 941 performs a plurality of statistical calculations to obtain at least one of a total value, an average value, and a median of a plurality of values y1 (n) to yN (n) constituting the frequency component of the plurality of feature quantities. Ask for at least one.

Therefore, the calculation unit 941 does not compare all of the plurality of values y1 (n) to yN (n) constituting the frequency component, but has a plurality of different representatives corresponding to a plurality of types of statistical calculations. Extract the values as multiple features.

Therefore, since the wire rope flaw detector uses a plurality of feature quantities as a plurality of representative values extracted from various aspects, the accuracy of determining the presence or absence of damage to the wire can be particularly significantly improved.

Embodiment 2.
In the second embodiment, the description of the same or equivalent configuration and function as the first embodiment is omitted. The second embodiment is different from the first embodiment in that either one of the learning model 961_1 and the learning model 961_2 of the first embodiment is already stored in the learning unit 943. Other configurations are the same as those in the first embodiment. That is, the other configurations are the same as or equivalent to those of the first embodiment, and these parts are designated by the same reference numerals.

FIG. 15 is a flowchart illustrating processing by the control unit 9 in the second embodiment. As described above, before the process of step S61 of FIG. 15 is started, either one of the learning model 961_1 and the learning model 961_2 is already stored in the learning unit 943. Therefore, when inspecting the wire rope 2S, the learning process including the processes of steps S11 to S13 as described with reference to FIG. 12 is unnecessary. Similarly, when inspecting the wire rope 2S, the learning process including the processes of steps S41 to S43 as described with reference to FIG. 14 is unnecessary. Further, in FIG. 15, the period determination process including the processes of steps S21 and S22 as described with reference to FIG. 12 is omitted. Similarly, in FIG. 15, the period determination process including the processes of steps S50 and S51 as described with reference to FIG. 14 is omitted.

Specifically, the processes of steps S61 and S62 are feature extraction processes. The processing of step S61 and step S62 is the same as the processing of step S44 and step S45. Therefore, the description thereof will be omitted.

Further, the processes of step S63 and steps S65 to S67 are wire line determination processes. The processes of steps S63 and steps S65 to S67 are the same as the processes of steps S46 to S49. Therefore, the description thereof will be omitted. Here, the processes of steps S65 to S67 are executed by the learning model 961_2 generated from the autoencoder 943_2.

Further, the processes of step S63 and steps S68 to S71 are wire line determination processes. The processes of steps S63 and steps S68 to S71 are the same as the processes of steps S16 to S20. Therefore, the description thereof will be omitted. Here, the processes of steps S68 to S71 are executed by the learning model 961_1 generated from the support vector machine 943_1.

In step S64, the control unit 9 determines whether the learning model 961 is the learning model 961_2 generated by the autoencoder 943_2 or the learning model 961_1 generated by the support vector machine 943_1. When the control unit 9 determines that the learning model 961 is the learning model 961_2 generated by the autoencoder 943_2, the process of step S64 proceeds to the process of step S65. On the other hand, when the control unit 9 determines that the learning model 961 is the learning model 961_1 generated from the support vector machine 943_1, the process of step S64 proceeds to the process of step S68.

By using either the learning model 961_1 or the learning model 961_2 in this way, the sensor signal x (t) is input as an input to the control unit 9, and the wire is damaged as an output from the control unit 9. The judgment result of the presence or absence of is output.

Therefore, either one of the learning model 961_1 and the learning model 961_2 can be used for failure diagnosis of the wire rope 2S.

From the above description, since either one of the learning model 961_1 and the learning model 961_2 is stored in the learning unit 943, if the sensor signal x (t) is input to the control unit 9, the control unit 9 will wire the wire. The judgment result of the presence or absence of damage is output. Thereby, either one of the learning model 961_1 and the learning model 961_2 can be utilized for the failure diagnosis of the wire rope 2S.

Embodiment 3.
In the third embodiment, the description of the same or equivalent configuration and function as those of the first embodiment and the second embodiment is omitted. The third embodiment is different from the first and second embodiments in that the bandpass filters of the first and second embodiments are realized by wavelet transform. Other configurations are the same as those of the first embodiment and the second embodiment. That is, the other configurations are the same as or equivalent to those of the first embodiment, and these parts are designated by the same reference numerals.

FIG. 16 is a block diagram showing a functional configuration example of the control unit 9 that processes a signal corresponding to the leakage magnetic flux L_F leaking from the wire rope 2S according to the third embodiment. As shown in FIG. 16, the filter unit 193 uses a wavelet transform unit 1931 as a bandpass filter to generate a distribution of frequency components of the sensor signal x (t) by performing a wavelet transform on the sensor signal x (t). I have.

FIG. 17 is a diagram showing an example of the frequency characteristics of the filter unit 193 of FIG. The bandpass filter is realized by the basis function of the wavelet transform processed by the wavelet transform unit 1931. As shown in FIG. 17, each of the bandwidth b _k of the plurality of bands, the center frequency omega _ck bandwidth becomes narrower lower.

Next, as an example of the basis function of the wavelet transform, Mollet Wavelet will be described with reference to FIG. FIG. 18 is a diagram showing an example of a waveform in the time domain of the mother wavelet by the wavelet transform unit 1931 of FIG. The mother wavelet is expressed by the equation (1).

The daughter wavelet is expressed by the following equation (2). The scale of the daughter wavelet is expressed by the following equation (3). The daughter wavelet represented by the formula (2) can increase or decrease the amplitude of the waveform shown in FIG. 18 according to the scale represented by the formula (3). Further, the daughter wavelet represented by the equation (2) can translate the waveform shown in FIG. 18 in the time axis direction according to the scale represented by the equation (3). Here, s ₀ is a constant of scale. _sk is a scale function that takes k as an argument and _{is multiplied by s 0.}

Next, the Fourier transform of the mother wavelet and daughter wavelet is explained as follows. First, the Fourier transform equation of the mother wavelet is expressed by the following equation (4). On the other hand, the Fourier transform of the daughter wavelet is expressed in Eq. (5).

FIG. 19 is a diagram showing an example of waveforms in the frequency domain of the mother wavelet by the wavelet transform unit 1931 of FIG. As shown in FIG. 19, the frequency characteristic of Morlet Wavelet, among the frequency components of the input signal x (n), the frequency of the pass band specified by the center frequency omega ₀ of bandwidth b _k and bandwidth b _k It becomes a bandpass filter to pass. The center frequency ω _{ck in} FIG. 19 is expressed by the following equation (6). As expressed in equation (6), the center frequency ω _ck is _{expressed by the value obtained by dividing ω 0} / s ₀ by the power of the root of 2 to the power of m. Here, as explained above, m is a natural number.

Also, the bandwidth b _k in FIG. 19 is expressed by the following equation (7).

Here, the equations (6) and (7) in the case of m = 1 are explained as follows. First, the following equation (8) is an equation when m = 1 in the equation (6). From the description of equation (8), the center frequency ω _ck is expressed by the value obtained by dividing _{ω 0} / s _{0 by 2.}

On the other hand, the following equation (9) is the equation when m = 1 in the equation (7). From the description of equation (9), the bandwidth _bc is expressed by the value obtained by doubling the square root of the natural logarithm of 2 _{and dividing by s 0 by dividing by 2.}

Therefore, since Δk = 1 / m according to the equation (7), the equation (7) becomes Δk = 1 / m = 1/1 = 1. _{Therefore, the relationship between the bandwidth b k of} one of the two adjacent bands and the bandwidth b _{k + 1} of the other band having a lower center frequency ω _{ck than that of one band is b k +. The} ^{relationship of} 1 = 2-Δk · b _k = b _{k + 1} = 2 ^-1 · b _k is satisfied. Thus, for m = 1, the relationship between the bandwidth b _k, the bandwidth b _{k + 1} of the other band of one band is 1 octave.

Specifically, in the equation (8), when k = 0, the center frequency ω _ck becomes the frequency ω ₀ / s ₀ . Further, in the equation (9), when k = 0, the bandwidth _bc is expressed by the following equation (10). Therefore, from equations (8) and (9), for each k increases 1, the center frequency omega _ck and bandwidth b _k becomes 1/2.

Further, since Δk = 1 / m, the following relationship is established when m is a natural number other than 1. Each other among the two adjacent bands, the relationship between the bandwidth b _k of one band, the bandwidth b _{k + 1} of the other lower band center frequencies omega _ck than one band, b k _{+ 1} = ^{The relationship of} 2-Δk · b _k = b _{k + 1} = ^{2-1 / m} · b _k is satisfied. Thus, if m is not 1, the relationship between the bandwidth b _k, the bandwidth b _{k + 1} of the other band of one band becomes 1 / m octave. The center frequency ω _ck when m = 3 is explained as follows.

FIG. 20 is a conceptual diagram of _{the center frequency ω ck} at 1/3 octave as another example of the frequency characteristics of the filter unit 193 of FIG. As shown in FIG. 20, the center frequency ω _ck can be expressed by a value obtained by dividing ω ₀ / s ₀ by the power of the cube root of 2.

From the above description, the center frequency ω _ck and the center frequency ω _{ck + 1 of the} two bands adjacent to each other are expressed by the following equation (11). Equation (11) expresses the difference in magnitude between _{the center frequency ω ck} and the center frequency ω _{ck + 1 of} two bands adjacent to each other based on the equation (8). From equation (11), the center frequency ω _ck becomes ^{2-1 / m} each time k is incremented by 1.

_{Further, the bandwidth b k} and the bandwidth b _{k + 1} in each of the two adjacent bands are expressed by the following equation (12). Equation (12) expresses the difference in magnitude _{between the bandwidth b k} and the bandwidth b _{k + 1 of} two adjacent bands based on the equation (9). From equation (12), for each k increases 1, the bandwidth b _k will 2 ^{-1 / m.}

FIG. 21 is a diagram showing an example of the distribution of frequency components extracted from the signal corresponding to the leakage magnetic flux L_F by the filter unit 193 of FIG. An example of FIG. 21, except bandwidth b _k and the center frequency omega _ck two bands adjacent to each other, is similar to an example of FIG. 8. Therefore, the description of FIG. 21 is omitted.

FIG. 22 is a diagram showing an example of the sequence yk (n) generated by the filter unit 193 at the time t1 of FIG. In one example of FIG. 22, among a plurality of values y1 (n) to yN (n) constituting the sequence yk (n), the value y6 (n) when k = 6 indicates the maximum value, and k = 1, The values y1 (n), y2 (n), y3 (n), y5 (n) and y7 (n) when the values are 2, 3, 5 and 7 indicate the minimum values. Therefore, the range is obtained by the difference between the value y6 (n) and the values y1 (n), y2 (n), y3 (n), y5 (n), and y7 (n).

FIG. 23 is a diagram showing an example of the sequence yk (n) generated by the filter unit 193 at the time t2 of FIG. In one example of FIG. 23, the value located at the center when a plurality of values y1 (n) to yN (n) constituting the sequence yk (n) are arranged in ascending order is the value when k = 5. Therefore, the value when k = 5 is adopted as the median value.

FIG. 24 is a flowchart illustrating processing by the control unit 9 of FIG. The process of step S81 is a learning process. The processes of steps S82 to S84 are feature extraction processes. The process of step S85 is a wire determination process. The process of step S86 is a period determination process. Of the learning process, feature amount extraction process, wire determination process, and period determination process, the learning process, wire determination process, and period determination process are as follows. That is, the learning process, the wire line determination process, and the period determination process are a series of processes including the processes of steps S11 to S13, the processes of steps S16 to S20, and the processes of steps S21 and S22 in FIG. Further, the learning process, the wire line determination process, and the period determination process are a series of processes including the processes of steps S41 to S43, the processes of steps S46 to S49, and the processes of steps S50 and S51 in FIG. Therefore, the learning process, the wire line determination process, and the period determination process are a series of processes of any one of the above. Therefore, those explanations are omitted.

<Feature extraction process>
In step S82, the synthesizer 92 inputs the input signal x (n) corresponding to the sensor signal x (t) to the filter unit 193. The synthesizer 92 shifts the current process to the process of step S83.

In step S83, the filter unit 193 extracts the frequency component of the input signal x (n) by the wavelet transform unit 1931. The filter unit 193 shifts the current process to the process of step S84. The description of the process executed by the absolute value unit 932 between the process of step S83 and the process of step S84 will be omitted.

In step S84, the calculation unit 941 extracts a plurality of feature quantities from a plurality of values y1 (n) to yN (n) composed of frequency components extracted by the wavelet transform unit 1931 by a plurality of statistical calculations. The calculation unit 941 shifts the current processing to the processing in step S85.

From the above description, in the wire rope flaw detector, each of the bandwidth b _k of the plurality of bands, the center frequency omega _ck bandwidth becomes narrower lower. Therefore, the lower the center frequency ω _{ck of the} band, the higher the frequency resolution and the lower the time resolution. The higher the center frequency ω _{ck of the} band, the lower the frequency resolution and the higher the time resolution. Therefore, the wire rope flaw detector can more accurately detect where the sudden fluctuation occurred on the time axis, and can more accurately determine the frequency of the slow fluctuation, which enables efficient analysis.

_{Further, the relationship between the bandwidth b k of} one of the two adjacent _{bands and the bandwidth b k + 1} of the other band having a lower center frequency than one band is Δk = 1 / m. ^Then , the relationship of b _{k + 1} = 2-Δk · b _{k is satisfied.} Therefore, the band can be changed by the root of 2 m. Therefore, the temporal resolution is particularly remarkably improved in the high frequency region, and the spatial resolution is particularly remarkably improved in the low frequency region.

Further, the filter unit 193 extracts a frequency component from the sensor signal x (t) by executing a wavelet transform on the sensor signal x (t). Since the wavelet is a local function, there is a high correlation between the wavelet and the detection of the locally generated wire damage portion B_W. Therefore, the filter unit 193 can emphasize the damaged frequency component f_s among the frequency components. Therefore, the wire rope flaw detector can emphasize the frequency component of the induced voltage generated when the wire is damaged, so that the SN ratio can be particularly significantly improved.

Further, for each embodiment, the functions of each part of the wire rope flaw detector are realized by the processing circuit. That is, the wire rope flaw detector includes a processing circuit for executing the synthesizer 92, the filter unit 93, the filter unit 193, the calculation unit 941 and the learning unit 943. Even if the processing circuit is dedicated hardware, it is also called a CPU (Central Processing Unit, central processing unit, processing unit, arithmetic unit, microprocessor, microprocessor, processor, DSP) that executes a program stored in the memory. It may be.

FIG. 25 is a diagram illustrating a hardware configuration example. In FIG. 25, the processing circuit 201 is connected to the bus 202. When the processing circuit 201 is dedicated hardware, the processing circuit 201 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, an ASIC, an FPGA, or a combination thereof. Each of the functions of each part of the wire rope flaw detector may be realized by the processing circuit 201, or the functions of each part may be collectively realized by the processing circuit 201.

FIG. 26 is a diagram illustrating another hardware configuration example. In FIG. 26, the processor 203 and the memory 204 are connected to the bus 202. When the processing circuit is a CPU, the functions of each part of the wire rope flaw detector are realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in memory 204. The processing circuit realizes the functions of each part by reading and executing the program stored in the memory 204. That is, when the wire rope flaw detector is executed by the processing circuit, the steps of controlling the synthesizer 92, the filter unit 93, the filter unit 193, the calculation unit 941 and the learning unit 943 are eventually executed. It has a memory 204 for storing a program. Further, it can be said that these programs cause the computer to execute the procedure or method for executing the synthesizer 92, the filter unit 93, the filter unit 193, the calculation unit 941 and the learning unit 943. Here, the memory 204 includes, for example, a non-volatile or volatile semiconductor memory such as RAM, ROM, flash memory, EPROM, EEPROM, or a magnetic disk, flexible disk, optical disk, compact disk, mini disk, DVD, or the like. Applicable.

Note that some of the functions of each part of the wire rope flaw detector may be realized by dedicated hardware, and some of the functions may be realized by software or firmware. For example, the filter unit 93 and the filter unit 193 can realize their functions by a processing circuit as dedicated hardware. Further, the arithmetic unit 941 and the learning unit 943 can realize their functions by the processing circuit reading and executing the program stored in the memory 204.

In this way, the processing circuit can realize each of the above functions by hardware, software, firmware, or a combination thereof. Next, an example of realizing each of the above functions will be specifically described as follows.

FIG. 27 is a diagram showing a system configuration example in which at least one of the control units 9 of FIGS. 6 and 16 is incorporated into the terminal device 501 and used as a specific example of FIG. 25 or FIG. In the wire rope flaw detector, as shown in FIG. 27, the probe 1 detects damage to the wire rope 2S. The wire rope 2S, for example, suspends an elevator car. The wire rope 2S may be used for a crane.

The probe 1 detects damage to the wire while moving with respect to the wire rope 2S, for example, along a specific direction W_D. The probe 1 supplies, for example, a sensor signal x (t), which is an analog signal, to the AD converter 301 via a cable. The AD converter 301 converts an analog signal into a digital signal. The digital signal converted by the AD converter 301 is input to the terminal device 501. As the terminal device 501, for example, a personal computer is used. The terminal device 501 determines whether or not the wire is damaged by performing various signal processing on the digital signal input from the AD converter 301. In addition, the terminal device 501 displays the determination result of the presence or absence of damage to the wire.

FIG. 28 shows a system configuration example in which the processing content of the determination device 401 is supplied to the data logger 601 by incorporating at least one of the control units 9 of FIGS. 6 and 16 into the determination device 401 as a specific example of FIG. 25 or FIG. It is a figure which shows. The probe 1 supplies, for example, a sensor signal x (t) composed of an analog signal to the determination device 401 via a cable. The determination device 401 is equipped with a microcomputer. The determination device 401 is dedicated hardware. The determination device 401 converts an analog signal into a digital signal. The determination device 401 determines whether or not the wire is damaged by performing various signal processing on the converted digital signal. In addition, the determination device 401 notifies the determination result of the presence or absence of damage to the wire.

Note that the determination device 401 can supply various internally processed signals to an external device as analog signals or digital signals. As the external device, for example, a data logger 601 is used. The data logger 601 can display a waveform by inputting an analog signal or a digital signal from the determination device 401. Further, the data logger 601 can record the processing contents of the determination device 401.

FIG. 29 shows a system configuration in which at least one of the control units 9 of FIGS. 6 and 16 is incorporated into the determination device 401 as a specific example of FIG. 25 or 26 to supply the processing contents of the determination device 401 to the elevator control panel 701. It is a figure which shows an example. By inputting a digital signal from the determination device 401, the elevator control panel 701 can transmit monitoring information such as which wire rope 2 of which property is broken to the central monitoring center.

As described above, the wire rope flaw detector has been described based on the first and second embodiments, but the present invention is not limited to this.

In the first and second embodiments, an example in which the learning unit 943 calculates a learning model 961 applicable to other data and then determines the presence or absence of a strand in the target data has been described, but the present invention is particularly limited to this. It is not something that is done. For example, the learning unit 943 may determine the presence or absence of a wire by the Mahalanobis distance MD after defining the unit space by the calculation of the MT method.

<MT method>
Specifically, the learning unit 943 defines a unit space when the MT method is used. The learning unit 943 calculates the distance from the center of the unit space to the target data as the Mahalanobis distance MD. The learning unit 943 determines that the Mahalanobis distance MD is short as normal. On the other hand, the learning unit 943 determines that the Mahalanobis distance MD is long as abnormal.

More specifically, in the learning unit 943, when there is no damage to the wire, the sensor signal x (t) and a plurality of values y1 (n) to yN ( The presence or absence of damage to the wire is determined using the correlation with n).

First, the calculation unit 941 calculates a standardized value obtained by standardizing each of the sensor signal x (t) and the plurality of values y1 (n) to yN (n). The calculation unit 941 calculates the standardized value by (raw data-average value) / standard deviation. Next, the calculation unit 941 calculates the correlation matrix R with the set of standardized values as the unit space. Next, the calculation unit 941 calculates the inverse matrix R ^-1 of the correlation matrix R. The calculated inverse matrix R ^-1 is held by the learning unit 943.

Next, the calculation unit 941 includes a sensor signal x (t) of the target data to be subject to the presence or absence of damage to the wire, and a plurality of values y1 (n) to which constitute a frequency component of the sensor signal x (t). A matrix Y is generated in which each of yN (n) is standardized in the same manner as described above. Next, the learning unit 943 ^{multiplies the matrix Y that standardizes the target data, the inverse matrix R -1 calculated in advance, and the transposed matrix Y T} of the matrix Y that standardizes the target data to obtain the number of items. By dividing, the Mahalanobis distance MD is calculated. The learning unit 943 determines whether or not the wire is damaged by comparing the preset threshold distance with the Mahalanobis distance MD. Here, the number of items is the number of parameters used in the standardization calculation, and in the above example, it is 1 + N.

That is, the control unit 9 uses the correlation between various data to derive the correspondence between the input data and the presence or absence of damage to the strands. As a result, the control unit 9 can reliably determine the presence or absence of damage to the strands based on the derived correspondence even when other input data is input. Therefore, the wire rope flaw detector can achieve a high generalization ability.

2,2S wire rope, 11 magnetometer, 13 magnetic sensor, 9 control unit, 93,193 filter unit, 94 processing unit, 941 arithmetic unit, 943 learning unit, 943_1 support vector machine, 943_2 autoencoder, 961,961_1,961_2 Learning model.

Claims (11)

1. A magnetizer that generates magnetic flux that passes through a part of the wire rope,
   A magnetic sensor that generates a signal corresponding to the leakage magnetic flux leaking from the wire rope as a sensor signal among the magnetic fluxes,
   A control unit that processes the sensor signal and
   With
   The control unit
   A filter unit that extracts the frequency component of the sensor signal and
   An arithmetic unit that extracts a plurality of features based on a plurality of values constituting the frequency component, and a calculation unit.
   Based on a plurality of values constituting the frequency component extracted by the filter unit, the trained learning model obtained by machine learning about the correlation between the plurality of features and the state of the wire rope contained in the wire rope is used. A learning unit that determines the presence or absence of damage to the wire rope by executing arithmetic processing by the learning model when a plurality of feature quantities extracted by the arithmetic unit are input.
   Has a wire rope flaw detector.
2. The calculation unit extracts the plurality of features by a plurality of statistical calculations, and obtains the plurality of features.
   The learning unit sets at least one of a plurality of first feature quantities that characterize the presence of damage to the strands and a plurality of second feature quantities that characterize the absence of damage to the strands for learning. The wire rope flaw detector according to claim 1, wherein the presence or absence of damage to the wire is determined from the output of the learning model.
3. The learning unit can classify the preset feature space into a first region including the plurality of first feature quantities and a second region including the plurality of second feature quantities by an identification hyperplane. Supported as a support vector machine,
   The support vector machine maps the plurality of first feature quantities and the plurality of second feature quantities to the feature space as the training data set and calculates the identification hyperplane to obtain the learning model. The wire rope flaw detector according to claim 2, which is generated.
4. The support vector machine determines that the wire is damaged when the output of the learning model when the plurality of features are input to the learning model corresponds to those classified into the first region. The third aspect of claim 3, wherein when the output of the learning model when the plurality of feature quantities are input to the learning model corresponds to those classified into the second region, it is determined that there is no damage to the wire rope. Wire rope flaw detector.
5. The learning unit is configured as an autoencoder that suppresses the error between input and output to a certain amount by a weighting coefficient.
   2. The autoencoder generates the learning model by calculating the weighting coefficient by using the plurality of second feature quantities as the learning data set for each of the input and the output of the autoencoder. The wire rope flaw detector described in.
6. When the output of the learning model when the plurality of features are input to the learning model can reconstruct the plurality of features, the autoencoder determines that there is no damage to the wire rope, and determines that the plurality of features are not damaged. The wire rope flaw detection according to claim 5, wherein when the output of the learning model when the feature amount of the above is input to the learning model cannot reconstruct the plurality of feature amounts, it is determined that the wire is damaged. Device.
7. The wire rope flaw detector according to any one of claims 2 to 6, wherein the filter unit has a plurality of bandpass filters having a plurality of bands different from each other as individual pass bands.
8. The wire rope flaw detector according to any one of claims 2 to 6, wherein the bandwidth of each of the plurality of bands becomes narrower as the center frequency of the band becomes lower.
9. Of the adjacent two of the bands, and bandwidth b _k of the one band, the relationship between the bandwidth b _{k + 1} of the other band lower center frequency than the band of the one is, m is a natural number, If Δk = 1 / m,
   ^{_{b k + 1 = 2 -Δk ·}} b k wire rope flaw detector according related to claim 8 meets the.
10. The wire rope flaw detector according to claim 8 or 9, wherein the filter unit extracts the frequency component from the sensor signal by performing a wavelet transform on the sensor signal.
11. The calculation unit obtains at least one of the total value, the average value, and the median value of the plurality of values as at least one of the plurality of feature quantities by the plurality of statistical calculations, according to claims 2 to 10. The wire rope flaw detector according to any one item.

Images