A method and system for identifying damage of traction machine wire rope based on Monte Carlo algorithm

Abstract

The present invention relates to the technical field of wire rope damage detection, and discloses a method and system for identifying damage to a traction machine wire rope based on a Monte Carlo algorithm, including an offline training step and an online detection step, wherein the offline training step includes: S10, taking wire ropes with damage notches of different specifications as training samples, collecting corresponding magnetic leakage signals after deep saturation magnetization of the wire ropes, and taking the characteristic values ​​of the magnetic leakage signals as input layer parameters, and taking the damage specifications as output layer parameters, training a neural network to obtain a trained neural network. The beneficial effects are: compared with manual detection, the present invention is based on detection data, quantitatively outputs the damage specifications of the wire rope, has little safety hazard, and can reasonably utilize resources. Compared with BP neural network detection, the detection accuracy is higher, the recognition precision is higher, and it is more suitable for wire rope damage identification detection.

Description

Technical Field

[0001] The present invention relates to the technical field of wire rope damage detection, and in particular to a method and system for identifying wire rope damage of a traction machine based on a Monte Carlo algorithm. Background Art

[0002] Wire ropes are widely used in coal mining, metallurgy, transportation, petroleum and other fields. The safety and reliability of wire ropes during use are greatly affected by the damage and development trend during use. As the use time increases, various damages to the wire ropes are inevitable, which will reduce the performance of the wire ropes and seriously affect the normal and safe use of the wire ropes, so regular inspections are required.

[0003] In the prior art, some occasions still mainly rely on traditional manual observation methods to detect defects in wire ropes, and ensure the safe operation of wire ropes by regularly forcibly replacing wire ropes. However, the manual observation detection method lacks scientific and accurate quantitative detection results as a basis for judgment, which not only poses a safety hazard, but also results in a waste of resources.

[0004] The magnetic flux leakage detection method directly detects and evaluates the defects and damages of in-service wire ropes without changing the state and performance of the wire ropes, and then infers their breaking tension and service life. This method can detect all defects and damages of wire ropes. The conditions and distribution of various damages have different degrees of influence as the strength of the wire rope decreases, and the wire rope is often scrapped due to serious damage to a certain section. Therefore, it is particularly important to conduct timely and effective defect detection of wire ropes. In the prior art, magnetic flux leakage detection is often combined with BP neural network to improve the accuracy and effectiveness of the detection results. However, in actual application, the method of combining magnetic flux leakage detection with BP neural network still has the problems of untimely detection, low recognition accuracy during detection, and inability to identify the damage specifications of the wire rope. Summary of the invention

[0005] In order to overcome the problems in the prior art of manual detection of wire rope damage, which lacks quantitative data, wastes resources and poses safety hazards, and the method combining magnetic flux leakage detection with BP neural network has untimely detection, low recognition accuracy and inability to identify the damage specifications of the wire rope, the present invention proposes a method and system for identifying damage to traction machine wire ropes based on Monte Carlo algorithm, and its detailed technical scheme is as follows:

[0006] A method for identifying damage of traction machine wire rope based on Monte Carlo algorithm includes an offline training step and an online detection step.

[0007] The offline training step includes: S10, taking steel wire ropes with damage notches of different specifications as training samples, collecting corresponding magnetic leakage signals after deep saturation magnetization of the steel wire ropes, and taking characteristic values ​​of the magnetic leakage signals as input layer parameters and taking damage specifications as output layer parameters to train the neural network and obtain a trained neural network;

[0008] The online detection steps include: S20, using a permanent magnet to deeply saturate the steel wire rope to be detected, and using a Hall sensor to obtain the leakage magnetic signal of the moving steel wire rope to be detected; S30, inputting the characteristic value of the leakage magnetic signal obtained in S20 into the trained neural network to obtain the damage specification of the corresponding steel wire rope output by the neural network.

[0009] Further, in S10 and S30, the characteristic values ​​are the peak value, valley value, peak-to-valley value, mean value, root mean square value and waveform area of ​​the magnetic flux leakage signal.

[0010] Further, in S10 and S30, the neural network is a Monte Carlo neural network, which is composed of three layers of neural networks: an input layer, a hidden layer, and an output layer;

[0011] Let M be the number of input layer nodes, H be the number of hidden layer nodes, and E be the number of output layer nodes, then:

[0012]

[0013] Among them, k h represents the local field of the hidden layer, z m represents the input layer unit; v hm represents the weight matrix linking the input layer; b h Represents the bias of the hidden layer, m=1,2,…,M.

[0014] Furthermore, in the Monte Carlo neural network, p h is the output of the hidden layer, t e is the output of the output layer, then:

[0015] p h =f zp (β h ·k h ), h=1,2,…,H (2)

[0016]

[0017] Among them, β h represents the gain factor of the hidden layer; f zp represents the activation function of the hidden layer; w eh represents the connection matrix connecting the hidden layer and the output layer; fpt It represents the activation function of the output layer.

[0018] Furthermore, the Monte Carlo neural network also includes a loss function:

[0019]

[0020] Among them, Q represents the number of training samples; g represents the separation interval, which is used to separate samples with different damage specifications; t e represents the actual output of the neural network; t e0 It represents the label of the neural network.

[0021] Furthermore, the training process of the Monte Carlo neural network includes the following steps: S101, initializing the magnetic flux leakage data; S102, providing the characteristic value of the magnetic flux leakage data as the input vector, and taking the damage specification of the wire rope as the target output vector; S103, calculating the output of each unit of the hidden layer and the output layer; S104, calculating the average recognition rate and recognition rate variance of the samples; S105, determining whether the average recognition rate of the samples meets the preset requirements, if so, the training ends, if not, randomly selecting the weights within the range for mutation, and repeating S103 and S104.

[0022] Furthermore, between S20 and S30, there is also included S201, normalizing the eigenvalues ​​of the magnetic flux leakage data so that the eigenvalues ​​are all within the range of [0, 1].

[0023] A traction machine wire rope damage identification system based on Monte Carlo algorithm is used to implement the above-mentioned traction machine wire rope damage identification method.

[0024] According to any of the above embodiments, the present invention has at least the following beneficial effects: the present invention proposes a method and system for identifying damage to a traction machine wire rope based on a Monte Carlo algorithm, introduces a Monte Carlo algorithm and randomly selects training samples to change weights in a neural network model, so that the neural network model has more hidden layers and higher recognition accuracy.

[0025] At the same time, in the selection of input parameters, the present invention uses the peak value, valley value, peak-to-valley value, mean value, root mean square value and waveform area of ​​the leakage magnetic signal as characteristic value input, and the neural network model outputs the damage specifications of the wire rope accordingly. Compared with manual detection, the present invention is based on the detection data and quantitatively outputs the damage specifications of the wire rope, with less safety hazards and reasonable use of resources. Compared with BP neural network detection, the detection accuracy is higher, the recognition precision is higher, and it is more suitable for wire rope damage identification and detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flow chart of the method of the present invention.

[0027] Figure 2 It is a waveform diagram of magnetic flux leakage data during the detection process of the present invention.

[0028] Figure 3 It is a schematic diagram of the training process of the neural network model based on the Monte Carlo algorithm of the present invention.

[0029] Figure 4 It is a structural schematic diagram of the Monte Carlo neural network algorithm structure of the present invention.

[0030] Figure 5 It is the ROC curve diagram corresponding to the steel wire ropes with different damage specifications of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] A method for identifying damage of traction machine wire rope based on Monte Carlo algorithm, such as Figure 1 As shown, it includes an offline training step of establishing a Monte Carlo neural network model and an online detection step of using the Monte Carlo neural network model for wire rope damage identification, which specifically includes the following steps.

[0033] The offline training step includes the following steps:

[0034] 1. Obtain training samples. Artificially manufacture steel wire ropes with different damage specifications, where the damage specifications refer to the damage depth and damage width. In a specific embodiment of the present invention, a steel wire rope with a diameter of 10 mm is used, and an angle grinder is used to artificially manufacture damage notches on the steel wire rope. Five different damage specifications of notches are set in terms of damage depth and damage width: 2mm×2mm, 2mm×4mm, 2mm×6mm, 2mm×8mm, and 2mm×10mm.

[0035] The purpose of setting different damage specifications is to increase the sample types, provide sufficient training set data, and reduce the overfitting problem. The overfitting problem refers to excellent performance in the process of learning the data model, but poor performance in actual verification and data analysis. Increasing the sample types and providing a large amount of training set data is the simplest and most effective way to reduce the overfitting problem.

[0036] 2. Deep saturation magnetization of the steel wire rope: In the electromagnetic detection process of the steel wire rope, magnetization is the first step to achieve detection. The magnetization of the steel wire rope is achieved by an excitation device. In the present invention, a permanent magnet is used as an excitation source. The permanent magnet adopts a rare earth permanent magnet with a high magnetic energy product because the rare earth permanent magnet has the advantages of high magnetic energy product, small size, light weight, and no need for power supply. When the diameter of the steel wire rope is increased, a multi-circuit magnetization method can be used to effectively increase the magnetic induction intensity of the steel wire rope within the measured range.

[0037] The reasons for saturation magnetization of the wire rope are as follows: the signal of a wire rope in a relaxed state may be larger during detection; for a wire rope in a tensioned state, the detected signal for a break of the same size will be smaller; due to hysteresis, the magnetic induction intensity is not single-valued before it is magnetized to the saturation state. In this case, when the detection sensor scans the same break in the forward and reverse directions, the detected signal will be inconsistent; during the detection process, when the wire rope moves faster, the magnetic field of the local magnetic field provided by the magnetizer will lag behind, causing the magnetization to weaken, and the change in the magnetization state caused by the change in the metal cross-sectional area of ​​the wire rope will affect the magnetic induction intensity in the wire rope, thereby affecting the detection accuracy.

[0038] 3. Collection of leakage magnetic data. During the wire rope damage detection process, the leakage magnetic data of the wire rope is collected through a wire rope detection instrument. In a specific embodiment of the present invention, a hand-held wire rope detection instrument is used to perform measurements during the movement of the wire rope. During this measurement process, jitter and collision will inevitably occur. When jitter and collision occur, the distance between the Hall sensor and the wire rope will change, causing regular deviations in the leakage magnetic signal on individual Hall sensors. However, this deviation will not affect the Hall sensor's collection of leakage magnetic signals, and wire rope damage can still be detected. The wire rope leakage magnetic signal collected by the detection instrument is as follows: Figure 2 shown.

[0039] 4. Establish a neural network model based on the Monte Carlo algorithm and train it with sample data. The training steps of the neural network model based on the Monte Carlo algorithm are as follows: Figure 3 As shown, it specifically includes: 1. Initializing the collected magnetic leakage data; 2. Providing input variables and target output vectors; 3. Calculating the output of each unit in the hidden layer and the output layer; 4. Calculating the average recognition rate and recognition rate variance of the samples; 5. Determining whether the recognition rate meets the requirements; 6. If not, randomly selecting weights within the range for mutation, and then repeating steps 4 and 5; 7. If the requirements are met, the training ends.

[0040] In order to improve the accuracy of machine learning, the selection of eigenvalues ​​of input parameters is extremely important. The selection of eigenvalues ​​determines the complexity of the model and the generalization ability of the training effect. In order to reduce the complexity of the model and improve the generalization ability of the model's training effect, it is necessary to select appropriate eigenvalues. If the eigenvalues ​​are not selected properly, it is easy to cause the model to overfit and reduce its performance. The leakage magnetic signal data of the wire rope is data that changes with time, and the selection of eigenvalues ​​can be determined according to its physical quantity. Since it is necessary to accurately calculate the distance and similarity between the eigenvalues, the present invention selects the following eigenvalues ​​based on the structure of this data: (1) Peak value: the highest value of the waveform in the time domain period, reflecting the maximum amplitude of the signal; (2) Valley value: the lowest value of the waveform in the time domain period, reflecting the minimum amplitude of the signal; (3) Peak-to-valley value: the difference between the peak value and the valley value, reflecting the amplitude range of the signal; (4) Mean value: the average value of the waveform in the time domain period, reflecting the DC component of the signal; (5) Root mean square value: the square root of the mean of the sum of squares of each data sample point, reflecting the effective value of the signal; (6) Waveform area: the integral area of ​​the signal on the time axis, reflecting the energy of the signal.

[0041] In order to improve the accuracy, 50 sets of data were collected for the wire rope, 10 sets of magnetic flux leakage data were collected for each damage specification, and all the data were normalized. The normalization formula is:

[0042]

[0043] Normalizing the data is to use a linear function to convert the original data into the range of [0,1]. This can effectively reduce the problems of large differences in numerical ranges and non-convergence of training caused by different characteristics of sample data. After normalization, all data are easy to be processed in the same value space, making the training model more unified and standardized.

[0044] The Monte Carlo neural network algorithm also consists of a basic three-layer neural network, namely the input layer, hidden layer and output layer. The structure of the Monte Carlo neural network algorithm is as follows Figure 4 shown.

[0045] Let M be the number of input layer nodes, H be the number of hidden layer nodes, and E be the number of output layer nodes, then:

[0046]

[0047] Among them, k h represents the local field of the hidden layer, z m (m=1,2,…,M) represents the input layer unit; v h x represents the weight matrix connecting the input layer; b h represents the bias of the hidden layer. The output p of the hidden layer h(h=1,2,…,H) and the output t of the output layer e (e=1,2,…,E) is:

[0048] p h =f zp (β h ·k h ), h=1,2,…,H (2)

[0049]

[0050] Among them, β h represents the gain factor of the hidden layer, which controls the smoothness of the neural network; f zp represents the activation function of the hidden layer; w eh represents the connection matrix connecting the hidden layer and the output layer; f pt It represents the activation function of the output layer. Since the present invention is mainly used to identify wire rope damage, the following loss function is defined:

[0051]

[0052] Where Q represents the number of samples; g represents the separation interval, which is mainly used to separate samples with different damage specifications; t e represents the actual output of the neural network; t e0 It represents the label of the neural network. In the Monte Carlo neural network model, the output of each network is the sample recognition rate. The average sample recognition rate is obtained by averaging different sample recognition rates. After repeated model calculations, the variance of the output result, that is, the variance of the sample recognition rate, is minimized, which meets the training requirements.

[0053] The normalized peak value (Z1), valley value (Z2), peak-to-valley value (Z3), mean value (Z4), waveform area (Z5), and root mean square value (Z6) feature values ​​are used as the input of the model. 2mm (t1), 4mm (t2), 6mm (t3), 8mm (t4), and 10mm (t5) are used as the output of the model. In order to improve the accuracy of the Monte Carlo neural network algorithm in identifying wire rope damage, it is necessary to increase the number of training sets of the model to increase the samples and general vectors that can be randomly selected in the model. This paper uses 80% of the 50 groups of experimental samples as the training set of the model, and the remaining 20% ​​of the samples are used as the test set of the model. After the Monte Carlo neural network algorithm model is established, the test samples are input into it multiple times. Through certain parameter adjustments, the optimal training result accuracy is 86.4%. The ROC curves corresponding to different wire rope damage specifications are shown below. Figure 5 shown.

[0054] The area under the ROC curve (AUC) is an important criterion. The larger the AUC value, the better the model effect. For wire ropes with 2mm, 4mm, 6mm, 8mm, and 10mm damage, the AUCs corresponding to the results output by the Monte Carlo neural network algorithm model are 0.91, 0.87, 0.83, 0.85, and 0.86 respectively.

[0055] The Monte Carlo neural network algorithm of the present invention can effectively identify different damage defects of wire ropes, especially for the defects of 2mm, 4mm and 10mm in the experimental data. The leakage magnetic signals of these three types of damage have more obvious characteristics than the other two types. On the one hand, the leakage magnetic signal of the wire rope will decrease with the increase of the gap width. 2mm and 4mm can produce stronger leakage magnetic signals than other gaps, and are more significant than 6mm, 8mm and 10mm gaps in signal characteristics; on the other hand, the leakage magnetic signal wave width of the 10mm width gap is longer than that of other gaps. This feature makes the 10mm wire rope damage easier to be identified; therefore, in terms of recognition, 2mm, 4mm and 10mm wire rope damage can be more easily identified than 6mm and 8mm damage.

[0056] Compared with the BP neural network in the prior art, the Monte Carlo neural network has more hidden layers and higher recognition accuracy because it introduces the Monte Carlo algorithm and randomly selects training samples to change weights. The accuracy of using the BP neural network to identify the above wire rope damage is only 74.6%, while the accuracy of the Monte Carlo neural network algorithm is 86.4%. Therefore, the Monte Carlo neural network algorithm is more suitable for wire rope damage identification.

[0057] The online detection steps include: S20, using a permanent magnet to deeply saturate the steel wire rope to be detected, and using a Hall sensor to obtain the leakage magnetic signal of the moving steel wire rope to be detected; S30, inputting the characteristic value of the leakage magnetic signal obtained in S20 into the trained neural network to obtain the damage specification of the corresponding steel wire rope output by the neural network.

[0058] A traction machine wire rope damage identification system based on Monte Carlo algorithm is used to implement the above-mentioned traction machine wire rope damage identification method.

[0059] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying damage to a traction machine wire rope based on a Monte Carlo algorithm, comprising an offline training step and an online detection step, characterized in that: The offline training steps include: S10, using steel wire ropes with damage notches of different specifications as training samples, collecting corresponding magnetic leakage signals after deep saturation magnetization of the steel wire ropes, and using characteristic values ​​of the magnetic leakage signals as input layer parameters and damage specifications as output layer parameters to train the neural network and obtain a trained neural network; The online detection steps include: S20, using a permanent magnet to deeply saturate the steel wire rope to be tested, and using a Hall sensor to obtain a leakage magnetic signal of the moving steel wire rope to be tested; S30, inputting the characteristic value of the magnetic leakage signal obtained in S20 into the trained neural network to obtain the damage specification of the corresponding wire rope output by the neural network; In S10 and S30, the neural network is a Monte Carlo neural network, which is composed of three layers of neural networks: an input layer, a hidden layer, and an output layer; Let M be the number of input layer nodes, H be the number of hidden layer nodes, and E be the number of output layer nodes, then: Among them, k h represents the local field of the hidden layer, z m represents the input layer unit; v hm represents the weight matrix linking the input layer; b h represents the bias of the hidden layer, m=1,2,…,M; In the Monte Carlo neural network, p h is the output of the hidden layer, t e is the output of the output layer, then: p h =f zp (β h ·k h ),h=1,2,…,H (2) Among them, β h represents the gain factor of the hidden layer; f zp represents the activation function of the hidden layer; w eh represents the connection matrix connecting the hidden layer and the output layer; f pt It represents the activation function of the output layer.

2. The method for identifying damage to a traction machine wire rope based on a Monte Carlo algorithm according to claim 1, characterized in that: In S10 and S30, the characteristic values ​​are the peak value, valley value, peak-to-valley value, mean value, root mean square value and waveform area of ​​the magnetic flux leakage signal.

3. The method for identifying damage to a traction machine wire rope based on a Monte Carlo algorithm according to claim 1, characterized in that: The Monte Carlo neural network also includes a loss function: Among them, Q represents the number of training samples; g represents the separation interval, which is used to separate samples with different damage specifications; t e represents the actual output of the neural network; t e0 It represents the label of the neural network.

4. The method for identifying damage to a traction machine wire rope based on a Monte Carlo algorithm according to claim 3, characterized in that: The training process of the Monte Carlo neural network includes the following steps: S101, initializing magnetic flux leakage data; S102, providing the characteristic value of the magnetic flux leakage data as an input vector, and taking the damage specification of the wire rope as a target output vector; S103, calculating the output of each unit of the hidden layer and the output layer; S104, calculating the average recognition rate and recognition rate variance of the samples; S105, determining whether the average recognition rate of the samples meets the preset requirements, if so, the training ends, if not, randomly selecting weights within the range for mutation, and repeating S103 and S104.

5. The method for identifying damage to a traction machine wire rope based on a Monte Carlo algorithm according to claim 1, characterized in that: Between S20 and S30, S201 is also included, in which the eigenvalues ​​of the magnetic flux leakage data are normalized so that the eigenvalues ​​are all within the range of [0, 1].

6. A traction machine wire rope damage identification system based on Monte Carlo algorithm, characterized in that: Used to implement the method for identifying damage to a traction machine wire rope as described in any one of claims 1-5.

Images