A belt conveyor roller abnormality detection system based on optical fiber auscultation

Abstract

The present invention belongs to the technical field of mine transportation, and discloses a belt conveyor roller abnormality detection system based on optical fiber auscultation, comprising: a data acquisition unit, a preprocessing unit and a prediction model connected in sequence, and an abnormality detection unit and a linear layer connected to the prediction model. The data acquisition unit collects audio data during the operation of the roller, converts it into vibration data, and outputs it to the preprocessing unit. The preprocessing unit segments the input vibration data, and inputs the segmented vibration data into a trained prediction model. The prediction model outputs the features of the latent space corresponding to the current vibration data to the abnormality detection unit, and calculates the distance between the features of the current latent space and the feature center of the latent space when the roller is operating normally. If the distance is greater than the set distance threshold, the corresponding segment of the roller is currently operating abnormally. Not only is the automatic detection of roller abnormalities realized, but the latent space features output by the prediction model contain richer information, thereby improving the accuracy of roller abnormality detection.

Description

Technical Field

[0001] The invention belongs to the technical field of mine transportation, and more specifically, the invention relates to a belt conveyor roller abnormality detection system based on optical fiber auscultation. Background Art

[0002] With the continuous improvement and perfection of the intelligent system of mines, the ore mining process usually uses a belt conveyor to transport the mined ore, especially underground. The driving device drives the conveyor belt carrying the ore to rotate. Rollers for bearing the weight are arranged at a certain interval under the conveyor belt. The rotation of the conveyor belt drives the rollers to rotate, which can reduce the loss of the conveyor belt caused by friction.

[0003] However, after the conveyor belt has been working at high load for a long time, the roller may malfunction or fail, such as stalling. The stalling of the roller causes sliding friction between the conveyor belt and the roller, generating high temperature or sparks, which may cause a fire. Therefore, timely detection of roller abnormalities is very important for production safety.

[0004] The commonly used method for detecting roller faults is mainly manual inspection, and its shortcomings are mainly reflected in: manual inspection requires manual inspection of belt conveyor equipment, and judging whether there is an abnormality based on the sound made by the rotation of the belt conveyor roller. This method requires a lot of manpower, low inspection efficiency, and the noise during the inspection process is harmful to the human body. Summary of the invention

[0005] The present invention provides a belt conveyor roller abnormality detection system based on optical fiber auscultation, aiming to improve the above-mentioned problem.

[0006] The present invention is implemented as follows: a belt conveyor roller abnormality detection system based on optical fiber auscultation, the system comprising:

[0007] A data acquisition unit, a preprocessing unit connected to the data acquisition unit, a prediction model connected to the preprocessing unit, an anomaly detection unit connected to the prediction model, and a linear layer;

[0008] The data acquisition unit collects audio data during the operation of the roller based on the auscultation optical fiber, converts the audio data into vibration data, and outputs the vibration data to the preprocessing unit. The preprocessing unit segments the input vibration data and inputs the segmented vibration data into the trained prediction model. The prediction model outputs the features of the latent space corresponding to the current vibration data to the anomaly detection unit. The anomaly detection unit calculates the distance between the features of the current latent space and the feature center of the latent space when the roller is operating normally. If the distance is greater than the set distance threshold, the corresponding segment of the roller is currently operating abnormally.

[0009] Furthermore, the prediction model training process is as follows:

[0010] The data acquisition unit collects audio data during the operation of the roller based on the auscultation optical fiber, converts the audio data into vibration data, and outputs the vibration data to the data processing unit; the data preprocessing unit extracts sample features and labels from the vibration data during the normal operation of the roller, constructs positive samples, and inputs the features of the positive samples into the prediction model. The prediction results output by the prediction model are transformed through a linear layer to obtain an output with the same dimension as the label.

[0011] Furthermore, the prediction model includes:

[0012] A time domain feature extraction module, a frequency domain feature extraction module, and a fusion module connected to the time domain feature extraction module and the frequency domain feature extraction module, wherein:

[0013] Time domain feature extraction module is used to extract vibration data The time domain characteristics of

[0014] Frequency domain feature extraction module is used to extract vibration data The frequency domain features are converted into the time domain to obtain the time domain features in the frequency domain;

[0015] The fusion module calculates the weights of the time domain features and the time domain features in the frequency domain, and fuses the time domain features and the time domain features in the frequency domain based on the weights to form vibration data. The corresponding latent space features.

[0016] Furthermore, the time domain feature extraction module uses k Transformer Encoder modules connected in sequence.

[0017] Furthermore, the frequency domain feature extraction module includes: a Fourier transform unit connected in sequence, a frequency unit formed by connecting a plurality of linear layers in sequence, and an inverse Fourier transform unit, wherein:

[0018] Fourier transform unit, calculates vibration data Each data segment The average , and calculate each data segment The corresponding average The difference is then Fourier transformed and the absolute value is taken The frequency amplitude is obtained, input into the frequency unit, and multiple linear transformations are performed through the linear layer, and then the inverse Fourier transform is performed through the inverse Fourier transform unit to obtain the transformed time domain features.

[0019] Furthermore, the fusion module calculates the weights of the time domain features and the time domain features in the frequency domain. The calculation process is as follows:

[0020] Calculate the fundamental frequency based on the frequency amplitude extracted by the Fourier transform unit k and its harmonic energy , and calculate the frequency domain energy , the weight of the time domain features in the frequency domain , time domain characteristics .

[0021] Furthermore, the process of obtaining the feature center of the latent space when the roller is operating normally is as follows:

[0022] After the prediction model is trained, the data acquisition unit collects the audio data during the normal operation of the roller, converts the audio data into vibration data, outputs the vibration data to the data processing unit, and the preprocessing unit segments the input current vibration data of the belt conveyor roller. The trained prediction model is input, and the prediction model outputs the features of the latent space corresponding to the current vibration data. The features of a specified number of latent spaces during the normal operation of the roller are obtained, and the mean of the features of the specified number of latent spaces is calculated. The features of the latent space corresponding to the mean are used as the feature center of the latent space during the normal operation of the roller.

[0023] Furthermore, the positive sample construction process is as follows:

[0024] When the roller is operating normally, the vibration data collected by the auscultation optical fiber is divided into data segments, the existence of each data segment Sample data;

[0025] After each data segment Sample data as training targets , taking the data segment as a feature, corresponding to the data segment behind The sampled data are used as labels to complete the construction of positive samples.

[0026] The belt conveyor roller abnormality detection system based on optical fiber auscultation provided by the present invention has the following beneficial technical effects:

[0027] The time domain and frequency domain information of the periodic data are fully utilized, and both the global features of the audio data and the local features are considered. Secondly, the abnormality detection of the roller can be realized by using only positive samples, avoiding the influence of insufficient negative samples of industrial equipment. Thirdly, not only the automatic detection of roller abnormalities is realized, but also the latent space features output by the prediction model contain richer information, which improves the accuracy of roller abnormality detection:

[0028] The conveyor belt in the mine generally has a long transportation distance and a large number of rollers, and the cost of using sensors to collect data is high. In addition, the working environment in the mine is harsh, the sensor stability is poor, and its installation has relatively high requirements for infrastructure. Based on this, the present invention chooses to place at least one section of stethoscope fiber along the transmission direction of the conveyor belt, which can not only greatly reduce the collection cost, but also better adapt to the harsh environment of the mine, and convert the sound data collected by the stethoscope fiber into a vibration signal as the original data.

[0029] Since industrial equipment is generally more durable and has less fault data, when applying the model to identify whether the belt conveyor roller is running normally, using normal samples to train the model can effectively avoid the problem of insufficient sample data. The present invention uses normal data as positive samples to find the range of positive sample features. When the current operation data of the belt conveyor roller exceeds the range, it is considered that the belt conveyor roller is running abnormally. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A structural block diagram of a belt conveyor roller abnormality detection system based on optical fiber auscultation provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The specific implementation modes of the present invention are further explained in detail below by describing the embodiments with reference to the accompanying drawings, so as to help those skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.

[0032] Figure 1 The structural block diagram of the belt conveyor roller abnormality detection system based on optical fiber auscultation provided by the embodiment of the present invention, for the convenience of description, the system includes:

[0033] A data acquisition unit; a preprocessing unit connected to the data processing unit, a prediction model connected to the data processing unit, an anomaly detection unit and a linear layer connected to the prediction model;

[0034] Among them, the data acquisition unit collects audio data during the operation of the roller based on the auscultation optical fiber, converts the audio data into vibration data, and outputs the vibration data to the data processing unit; in the training stage, the data preprocessing unit extracts sample features and labels from the vibration data of the roller during normal operation, constructs positive samples, and trains based on the positive sample prediction model; in the use stage, the data preprocessing unit extracts sample features from the vibration data of the current operation of the roller, and inputs the trained prediction model; the prediction model outputs the features of the latent space corresponding to the current vibration data, and outputs it to the anomaly detection unit. The anomaly detection unit detects whether the current vibration data is abnormal based on the feature range of the latent space corresponding to the positive sample, and outputs the anomaly judgment result or outputs the judgment result; the features of the latent space output by the prediction model are linearly transformed by the linear layer to obtain an output with the same dimension as the label.

[0035] (i) The training phase of the prediction model;

[0036] In the embodiment of the present invention, the positive sample construction process is specifically as follows:

[0037] (11) When the roller is operating normally, the vibration data collected by the auscultation fiber is divided into data segments, the existence of each data segment Sampling data, segmented vibration data It is expressed as follows:

[0038] ;

[0039] in, Indicates The first sampling data (vibration data).

[0040] (12) Replace the following data segment with Sample data as training targets , taking the above data segment as a feature, corresponding to the data segment behind The sampled data are used as labels to complete the construction process of the positive samples.

[0041] ;

[0042] In an embodiment of the present invention, the prediction model includes:

[0043] A time domain feature extraction module, a frequency domain feature extraction module, and a fusion module connected to the time domain feature extraction module and the frequency domain feature extraction module, wherein:

[0044] Time domain feature extraction module is used to extract vibration data The time domain characteristics of

[0045] Frequency domain feature extraction module is used to extract vibration data The frequency domain features are converted into the time domain to obtain the time domain features in the frequency domain;

[0046] The fusion module calculates the weights of the time domain features and the time domain features in the frequency domain, and fuses the time domain features and the time domain features in the frequency domain based on the weights to form vibration data. The corresponding latent space features.

[0047] In the embodiment of the present invention, the time domain feature extraction module uses k Transformer Encoder modules connected in sequence, and the Encoder module includes operations such as position encoding, multi-head attention, and multi-normalization.

[0048] Among them, the position encoding formula can be expressed as: , where the encoding matrix The element in row i and column j of The calculation method is: , Indicates dimension, which is a specified value; input vibration data Mapping through the fully connected layer, position encoding Add directly to the map to complete the vibration data The encoding.

[0049] In the embodiment of the present invention, the frequency domain feature extraction module includes: a Fourier transform unit connected in sequence, a frequency unit formed by connecting a plurality of linear layers in sequence, and an inverse Fourier transform unit, wherein:

[0050] Fourier transform unit, calculates vibration data Each data segment The average , and calculate each data segment The corresponding average The difference is then Fourier transformed and the absolute value is taken The frequency amplitude is obtained, input into the frequency unit, and multiple linear transformations are performed through the linear layer, and then the inverse Fourier transform is performed through the inverse Fourier transform unit to obtain the transformed time domain features.

[0051] In the embodiment of the present invention, the fusion module calculates the weights of the time domain features and the time domain features in the frequency domain, and the calculation process is as follows:

[0052] The weight calculation process of the time domain features in the frequency domain is as follows:

[0053] Calculate the fundamental frequency based on the frequency amplitude extracted by the Fourier transform unit k and its harmonic energy , and calculate the frequency domain energy , the calculation process is as follows:

[0054] ;

[0055] in, Indicates the first The frequency domain energy of the harmonics, is the total number of harmonics at the fundamental frequency k.

[0056] ;

[0057] in, It represents the first i Frequency amplitude, since the spectrum is symmetrical after Fourier transform, only half of it needs to be taken. Harmonic energy is the frequency domain energy part of i Only take k、 2k, ..., Mk .

[0058] Then the weight of the time domain feature in the frequency domain is , then the time domain characteristics .

[0059] The time domain features output by the time domain feature extraction module and the frequency domain feature extraction module and the time domain features in the frequency domain are weighted and summed, and the same dimension as the target output is obtained through linear layer transformation. The loss with the target is calculated and training is performed until the number of training times reaches the threshold or the loss converges. The loss function is the mean square loss, and the formula is: ;

[0060] in, is the label corresponding to the i-th positive sample, Indicates that the prediction model is for i The prediction results of the output of the positive sample features.

[0061] (ii) the use phase of the prediction model;

[0062] During the operation of the belt conveyor, the data acquisition unit collects audio data during the operation of the roller based on the auscultatory optical fiber, converts the audio data into vibration data, and outputs the vibration data to the data processing unit. The preprocessing unit segments the current vibration data of the input belt conveyor roller. The segmentation process is the same as step (11). The segmented vibration data The trained prediction model is input, and the prediction model outputs the features of the latent space corresponding to the current vibration data, and outputs it to the anomaly detection unit. The anomaly detection unit calculates the distance between the features of the current latent space and the feature center of the latent space when the roller is operating normally. If the distance is greater than the set distance threshold, it is determined that the current audio data deviates greatly from the audio data when the roller is operating normally, and the roller is currently operating abnormally.

[0063] In the embodiment of the present invention, the process of acquiring the feature center of the latent space during the normal operation of the roller is as follows: after the prediction model training is completed, the data acquisition unit collects the audio data during the normal operation of the roller, and converts the audio data into vibration data, outputs the vibration data to the data processing unit, and the preprocessing unit segments the current vibration data of the input belt conveyor roller, and the segmented vibration data is converted into the vibration data. The trained prediction model is input, and the prediction model outputs the features of the latent space corresponding to the current vibration data. Based on the features of a specified number of latent spaces obtained during the normal operation of the roller, the mean of the features of the specified number of latent spaces is calculated, and the feature of the latent space corresponding to the mean is used as the feature center of the latent space when the roller is operating normally.

[0064] The present invention has been described exemplarily. Obviously, the specific implementation of the present invention is not limited to the above-mentioned method. As long as various non-substantial improvements are made using the method concept and technical solution of the present invention, or the concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A belt conveyor roller abnormality detection system based on optical fiber auscultation, characterized in that: The system comprises: A data acquisition unit, a preprocessing unit connected to the data acquisition unit, a prediction model connected to the preprocessing unit, an anomaly detection unit connected to the prediction model, and a linear layer; The data acquisition unit collects audio data during the operation of the roller based on the auscultatory optical fiber, converts the audio data into vibration data, and outputs the vibration data to the preprocessing unit. The preprocessing unit segments the input vibration data, and inputs the segmented vibration data into the trained prediction model. The prediction model outputs the features of the latent space corresponding to the current vibration data to the anomaly detection unit. The anomaly detection unit calculates the distance between the features of the current latent space and the feature center of the latent space when the roller is operating normally. If the distance is greater than the set distance threshold, the corresponding segment of the roller is currently operating abnormally. The linear layer transforms the prediction results output by the prediction model during training, and the transformed prediction results have the same dimension as the label; Predictive models include: A time domain feature extraction module, a frequency domain feature extraction module, and a fusion module connected to the time domain feature extraction module and the frequency domain feature extraction module, wherein: The time domain feature extraction module is used to extract the time domain features of vibration data; The frequency domain feature extraction module is used to extract the frequency domain features of the vibration data and transform them into the time domain to obtain the time domain features in the frequency domain; The fusion module calculates the weights of the time domain features and the time domain features in the frequency domain, and fuses the time domain features and the time domain features in the frequency domain based on the weights to form the features of the latent space corresponding to the vibration data; The frequency domain feature extraction module includes: a Fourier transform unit connected in sequence, a frequency unit formed by connecting multiple linear layers in sequence, and an inverse Fourier transform unit, wherein: Fourier transform unit, calculates each data segment x in the vibration data X i The average , and calculate each data segment x i The corresponding average The difference is then Fourier transformed and the absolute value is taken to obtain the frequency amplitude, which is input into the frequency unit, and multiple linear transformations are performed through the linear layer, and then the inverse Fourier transform unit is used to perform inverse Fourier transform to obtain the transformed time domain features; Calculate the fundamental frequency k and its harmonic energy E based on the frequency amplitude extracted by the Fourier transform unit ff , and calculate the frequency domain energy E f , the weight w of the time domain feature in the frequency domain f =E ff / E f , the weight of the time domain feature w t =1-(E ff / E f ).

2. The belt conveyor roller abnormality detection system based on optical fiber auscultation according to claim 1 is characterized in that: The training process of the prediction model is as follows: The data acquisition unit collects audio data during the operation of the roller based on the auscultation optical fiber, converts the audio data into vibration data, and outputs the vibration data to the preprocessing unit; the data preprocessing unit extracts sample features and labels from the vibration data during the normal operation of the roller, constructs a positive sample, and inputs the features of the positive sample into the prediction model. The prediction result output by the prediction model is transformed by the linear layer to obtain an output with the same dimension as the label.

3. The belt conveyor roller abnormality detection system based on optical fiber auscultation according to claim 1 is characterized in that: The time domain feature extraction module uses k Transformer Encoder modules connected in sequence.

4. The belt conveyor roller abnormality detection system based on optical fiber auscultation according to claim 1 is characterized in that: The process of obtaining the feature center of the potential space when the roller is operating normally is as follows: After the prediction model is trained, the data acquisition unit collects audio data during the normal operation of the roller, converts the audio data into vibration data, and outputs the vibration data to the preprocessing unit. The preprocessing unit segments the current vibration data of the input belt conveyor roller, and inputs the segmented vibration data X into the trained prediction model. The prediction model outputs the features of the latent space corresponding to the current vibration data, obtains the features of the specified number of latent spaces during the normal operation of the roller, and calculates the mean of the features of the specified number of latent spaces. The features of the latent space corresponding to the mean are used as the feature center of the latent space when the roller is operating normally.

5. The belt conveyor roller abnormality detection system based on optical fiber auscultation according to claim 2 is characterized in that: The process of constructing positive samples is as follows: When the roller is operating normally, the vibration data collected by the auscultation optical fiber is divided into N data segments, and each data segment has l sampling data; The l′ sampled data after each data segment are used as the training target y, the data segment is used as a feature, and the l′ sampled data after the corresponding data segment are used as a label to complete the construction of the positive sample.

Images