Belt conveyor auscultation abnormity management method based on comparative learning and time sequence modeling

Abstract

The invention relates to the technical field of belt conveyor auscultation, and discloses a belt conveyor auscultation abnormity management method based on comparative learning and time sequence modeling, comprising the following steps: step 1, acquiring operation sound and equipment parameters, extracting frame features and acquiring a voiceprint baseline; 2, the carrier roller synchronization frequency is calculated, and period and sideband fingerprint prototype vectors are constructed; step 3, inputting the frame feature and the fingerprint prototype vector into a contrast learning network to obtain a contrast embedded vector and a time sequence feature vector; 4, the time sequence modeling network outputs three types of probabilities and classification original output values; step 5, calculating an abnormal score, generating an abnormal event and determining a grade; step 6, calculating a priority score and distributing a work order; and 7, backfilling the disposal information, calculating the total cost and storing the total cost. According to the invention, accurate detection, hierarchical management and operation and maintenance closed-loop processing of abnormal operation of the belt conveyor are realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of belt conveyor auscultation, and particularly relates to a belt conveyor auscultation abnormality management method based on contrast learning and time sequence modeling. BACKGROUND

[0002] Belt conveyors are widely used in bulk material transportation in industries such as coal mines, metallurgy and power generation, and the stability of their operation is directly related to the continuity and safety of production. In the long-term operation process, the belt conveyor is prone to abnormality due to faults such as roller wear, bearing damage or belt deviation, and if it is not discovered and handled in time, it will often lead to equipment downtime and cause great economic losses. Existing condition monitoring mainly relies on manual inspection or methods based on vibration signals. Manual inspection is limited by personnel experience and has problems such as low efficiency, strong subjectivity and late discovery. The method based on vibration signals needs to install a large number of sensors, which has high layout and maintenance costs, and the signal-to-noise ratio is low in a complex noisy environment, making it difficult to identify early abnormalities in time and accurately.

[0003] In recent years, sound signals have been gradually used for belt conveyor condition monitoring. However, existing acoustic methods mainly rely on simple feature threshold determination or traditional classification models, and do not fully utilize the time sequence correlation features and spectral modulation information in the operating sound. At the same time, due to the complexity of the belt conveyor operating environment, the load and belt speed often change, and the robustness of existing methods under different working conditions is insufficient, which is prone to false positives or false negatives. In addition, existing technologies usually only stay at the level of abnormality detection, lack of hierarchical management of abnormal events and effective linkage with operation and maintenance work orders and cost assessment, and are difficult to meet the actual needs of production sites for intelligent and refined management. SUMMARY

[0004] The application provides a belt conveyor auscultation abnormality management method based on contrast learning and time sequence modeling, which solves the technical problems of insufficient belt conveyor abnormality detection accuracy, poor cross-condition robustness, and lack of abnormality grading and operation and maintenance closed-loop management in related technologies.

[0005] The application provides a belt conveyor auscultation abnormality management method based on contrast learning and time sequence modeling, which includes the following steps:

[0006] Step 1, obtaining the operating sound data and device parameters of the belt conveyor, extracting frame features from the operating sound data, and obtaining a voiceprint baseline under normal working conditions, the device parameters including the roller diameter and the belt speed;

[0007] Step 2, calculating the roller synchronous frequency according to the device parameters, and constructing the cycle and sideband fingerprint prototype vector, wherein the cycle and sideband fingerprint prototype vector includes the modulation ratio and the kurtosis ratio;

[0008] Step 3, input the frame features and the period into the contrast learning network with the sideband fingerprint prototype vector, and perform contrast learning according to the set data enhancement and pairing rules to generate contrast embedding vectors and time sequence feature vectors;

[0009] Step 4, input the time sequence feature vector into the time sequence modeling network to output normal, mild abnormal and severe abnormal three types of probabilities and classification original output values;

[0010] Step 5, calculate the abnormal score according to the three types of probabilities, the classification original output values, the period and the sideband fingerprint prototype and the voiceprint baseline, and generate an abnormal event based on the abnormal score, calculate the event intensity, and determine the abnormal event level according to the set threshold;

[0011] Step 6, calculate the priority score according to the abnormal event level, the event intensity, the device criticality and the real-time load, and generate a work order and dispatch according to the set service level agreement;

[0012] Step 7, the maintenance personnel complete the on-site disposal according to the work order and backfill the disposal information, calculate the total cost according to the disposal information and the downtime, and store the disposal information and the total cost in the database.

[0013] Further, frame features are extracted from the running sound data, and a voiceprint baseline is obtained under normal working conditions, including:

[0014] Step 11, divide the running sound data into frames according to the first preset frame length and the second preset frame shift, perform fast Fourier transform on each frame, and calculate the mel frequency cepstral coefficient through the Mel filter bank, while extracting the instantaneous energy in each frame to obtain the combination of the mel frequency cepstral coefficient and the instantaneous energy;

[0015] Step 12, calculate the first-order difference and the second-order difference based on the mel frequency cepstral coefficient, and combine the mel frequency cepstral coefficient, the first-order difference, the second-order difference and the instantaneous energy into a frame feature set; wherein the first-order difference is obtained by the difference between adjacent frame features, and the second-order difference is obtained by the difference between adjacent first-order differences;

[0016] Step 13, collect continuous multiple frame feature sets under normal working conditions, wherein the frame feature set includes energy values of each frequency dimension obtained by the Mel filter bank; calculate the median of the energy value sequence of each frequency dimension as the reference energy level of the frequency, and calculate the difference between the ninety-fifth percentile value and the median of the frequency energy value sequence as the upper limit of the energy deviation of the frequency;

[0017] Step 14, combine each frequency's reference energy level and energy deviation upper limit into an ordered pair, and form a voiceprint baseline by collecting all ordered pairs.

[0018] Further, the device parameter is used to calculate the synchronous frequency of the roller, and a cycle and sideband fingerprint prototype is constructed, including:

[0019] In step 21, the circumference is calculated according to the diameter of the roller and the circumference, and the ratio of the belt speed to the circumference is taken as the synchronous frequency of the roller;

[0020] In step 22, the key frequency band is divided with the synchronous frequency of the roller as the center, and the frequency corresponding to the energy peak value is taken as the modulation frequency; the positive sideband and the negative sideband are respectively the synchronous frequency of the roller plus or minus the modulation frequency;

[0021] In step 23, the ratio of the sum of the energy of the frequency components in the positive sideband and the negative sideband to the sum of the energy of the frequency components in the main frequency band is calculated to obtain the modulation ratio, and the main frequency band is a range with a preset percentage above and below the synchronous frequency of the roller;

[0022] In step 24, the ratio of the kurtosis of the main frequency band signal to the kurtosis of the signal in the full frequency range is calculated to obtain the kurtosis ratio; wherein the kurtosis is the ratio of the fourth moment of the signal deviating from the mean value to the square of the variance;

[0023] In step 25, the modulation ratio and the kurtosis ratio are combined to form a cycle and sideband fingerprint prototype vector.

[0024] Further, compared with the generation of the embedding vector and the time sequence feature vector, including:

[0025] In step 31, the frame feature set is subjected to data enhancement processing, and the data enhancement includes time axis random movement and band-limited noise mixing;

[0026] In step 32, a paired sample is constructed based on the enhanced frame feature set and the cycle and sideband fingerprint prototype vector, wherein a positive sample pair is a feature sequence of the same device under the same working condition, and a negative sample pair is a feature sequence of different devices or different working conditions, and the working condition is determined according to the classification results of the load and the belt speed of the device;

[0027] In step 33, the paired sample is input into an encoder and a projector, the encoder includes a one-dimensional convolution layer and a gated recurrent unit, and is used to extract a time sequence feature vector, and the projector includes a multi-layer fully connected network, and is used to compress the time sequence feature vector to obtain a comparison embedding vector;

[0028] In step 34, a loss function is constructed and the parameters of the encoder and the projector are optimized, the loss function includes a comparison loss and a cycle regularization loss, and is combined by weighting through a preset weight, wherein the comparison loss is calculated based on the similarity of the comparison embedding vectors of the positive sample pair and the similarity of the comparison embedding vectors of the negative sample pair, and the cycle regularization loss is calculated based on the time sequence feature vectors of the same signal before and after phase shift;

[0029] Step 35, the encoder processed in step 34 is used with the projector for forward calculation, and the contrast embedding vector and the time sequence feature vector are output as the input of the time sequence modeling network.

[0030] Further, the data enhancement further includes frequency interval masking and time frame masking, wherein the frequency interval masking refers to randomly selecting a continuous frequency interval on the frequency dimension of the frame feature set and setting the features of the interval to zero, and the time frame masking refers to randomly selecting a continuous frame on the time dimension of the frame feature set and setting the part of the features to zero.

[0031] Further, the generation of the three types of probabilities includes:

[0032] Step 41, the optimized time sequence feature vector is segmented and processed, and the time sequence feature vectors of continuous frames are divided into segment-level feature sequences according to a preset length;

[0033] Step 42, the segment-level feature sequences are input into a branch network for decomposition, wherein a low-frequency trend branch extracts long-term dependence features through a gated recurrent unit, a high-frequency impact branch extracts local impulse features through convolution operation and a gating mechanism, and the long-term dependence features and the local impulse features are spliced in the same feature dimension to form an enhanced feature vector;

[0034] Step 43, the enhanced feature vector is input into a classifier, and a classification original output value vector containing normal, mild abnormality and severe abnormality is output, and a three-type probability is obtained through normalization calculation of a softmax function.

[0035] Further, the calculation of the abnormal score and the determination of the abnormal event level include:

[0036] Step 51, subtracting the normal probability from the classification contribution, taking the part greater than zero in the square sum difference of the classification original output value vectors of adjacent two segments as the spectral flux contribution, taking the modulation ratio as the sideband contribution, and taking the weighted sum of the three parts according to a preset weight to obtain an abnormal score;

[0037] Step 52, based on the abnormal score sequence of continuous segments, a centralized abnormal score is calculated by subtracting the upper limit of the abnormal score, and a one-sided cumulative sum is performed on the centralized abnormal score, and when the one-sided cumulative sum exceeds a preset deviation threshold and lasts for more than three continuous segments, an abnormal event is determined;

[0038] Step 53, in the window corresponding to the abnormal event, the abnormal score is first subjected to exponential transformation, then the transformed score is averaged and taken as a natural logarithm to obtain an event intensity;

[0039] Step 54, determining as normal when the event intensity is lower than the first threshold, determining as mild abnormal when the event intensity is between the first threshold and the second threshold, and determining as severe abnormal when the event intensity is not lower than the second threshold.

[0040] Further, the step 6 specifically comprises:

[0041] Step 61, acquiring the abnormal event level, the event intensity, the equipment criticality and the real-time load, and obtaining the priority score through weighted summation;

[0042] Step 62, performing service level agreement mapping according to the priority score and the abnormal event level;

[0043] Step 63, under the premise of meeting the skill matching, dispatching the generated work order to the maintenance personnel, the skill matching is judged based on the correspondence between the required skill label of the work order and the skill label of the maintenance personnel, and the shortest path distance between the work order location and the current location of the maintenance personnel is used as the dispatch priority condition.

[0044] Further, the treatment information includes treatment time, material consumption, replacement component code and treatment result;

[0045] The equipment recovery operation time is determined according to the treatment time of the treatment information, and the downtime length is obtained in combination with the downtime start time;

[0046] The total cost includes: the fixed cost corresponding to the work order type, the material cost and the downtime cost, and the downtime cost is obtained by multiplying the downtime loss rate per unit time and the downtime length.

[0047] The beneficial effects of the present application are that: the present application introduces the construction of frame features and voiceprint baseline at the feature level, which can accurately reflect the normal operation acoustic characteristics of the equipment; by establishing the cycle and sideband fingerprint prototype vector, the equipment operation mechanism and the spectral characteristics are combined, the physical rationality and interpretability of the feature representation are improved; at the modeling level, the contrast learning is used to realize the robust representation across working conditions, and through the decoupling fusion of the low-frequency trend branch and the high-frequency impact branch, the dual sensitivity of the model to trend anomalies and sudden anomalies is enhanced; at the decision level, through the hierarchical calculation of the abnormal score, the abnormal event and the event intensity, quantitative identification and grading determination of the abnormality can be realized; further, the priority score is calculated by combining the equipment criticality, the real-time load and the production schedule, which ensures the rationality of the work order dispatch and the timeliness of the response. Overall, the present application realizes the closed-loop management from the belt conveyor sound signal acquisition, the abnormality detection to the operation and maintenance treatment and the cost archiving, which can significantly improve the intelligentization and the fine level of the belt conveyor operation and maintenance, and reduce the downtime risk and the maintenance cost. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1This is a flowchart of the belt conveyor auscultation anomaly management method based on contrastive learning and temporal modeling of the present invention. Detailed Implementation

[0049] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0050] like Figure 1 As shown, the belt conveyor auscultation anomaly management method based on contrastive learning and temporal modeling includes the following steps:

[0051] Step 1: Obtain the operating sound data and equipment parameters of the belt conveyor; extract frame features based on the operating sound data; and obtain the acoustic baseline under normal operating conditions. The equipment parameters include the idler diameter and belt speed.

[0052] Step 2: Calculate the roller synchronization frequency based on the equipment parameters, and construct the period and side-band fingerprint prototype vector, where the period and side-band fingerprint prototype vector includes the modulation ratio and kurtosis ratio;

[0053] Step 3: Input the frame features and periodicity with the sideband fingerprint prototype vector into the contrast learning network, perform contrast learning according to the set data augmentation and pairing rules, and generate contrast embedding vectors and temporal feature vectors;

[0054] Step 4: Input the temporal feature vector into the temporal modeling network, and output the probabilities of three classes: normal, mildly abnormal, and severely abnormal, as well as the original classification output values.

[0055] Step 5: Calculate the anomaly score based on the three probabilities, the original output value of the classification, the period and the side-band fingerprint prototype and the voiceprint baseline, generate anomaly events based on the anomaly scores, calculate the event intensity, and determine the anomaly event level according to the set threshold.

[0056] Step 6: Calculate the priority score based on the abnormal event level, event intensity, equipment criticality, and real-time load, and generate and dispatch work orders according to the set service level agreement.

[0057] Step 7: The maintenance personnel complete the on-site handling according to the work order and fill in the handling information. They calculate the total cost based on the handling information and downtime, and store the handling information and total cost in the database.

[0058] In one embodiment of the present invention, the operating sound data refers to the original sound signal collected by a microphone during the normal operation of the belt conveyor;

[0059] Frame features are extracted according to the running sound data, and a voiceprint baseline is obtained under normal working conditions, comprising:

[0060] Step 11, frame division is performed on the running sound data according to a first preset frame length and a second preset frame shift, fast Fourier transform is performed on each frame, and mel frequency cepstral coefficients are calculated through a mel filter bank, while instantaneous energy is extracted in each frame to obtain a combination of mel frequency cepstral coefficients and instantaneous energy; wherein the instantaneous energy refers to the amplitude square sum of each frame signal, which can reflect the intensity of the signal in the time segment;

[0061] Step 12, first-order difference and second-order difference are calculated based on the mel frequency cepstral coefficients, and the mel frequency cepstral coefficients, the first-order difference, the second-order difference and the instantaneous energy are combined into a frame feature set; wherein the first-order difference is obtained by the difference between adjacent frame features, and the second-order difference is obtained by the difference between adjacent first-order differences;

[0062] Step 13, a plurality of frame feature sets are collected under normal working conditions, wherein the frame feature set includes energy values of each frequency dimension obtained by the mel filter bank; the median of the energy value sequence of each frequency dimension is calculated as the reference energy level of the frequency, and the difference between the ninety-fifth percentile value and the median of the frequency energy value sequence is calculated as the upper limit of the energy deviation of the frequency;

[0063] Step 14, the reference energy level of each frequency and the upper limit of the energy deviation are combined into an ordered pair, and all ordered pairs are combined to form a voiceprint baseline, which is used to represent the statistical feature set of the sound signal of the belt conveyor under normal working conditions, and provides a reference for subsequent anomaly detection.

[0064] In an embodiment of the present application, the synchronous frequency of the roller is calculated according to the device parameters, and the period and the sideband fingerprint prototype are constructed, comprising:

[0065] Step 21, the circumference is calculated according to the diameter of the roller and the circumference, and the ratio of the belt speed to the circumference is taken as the synchronous frequency of the roller;

[0066] Step 22, the key frequency band is divided with the synchronous frequency of the roller as the center, and the frequency corresponding to the energy peak value is selected as the modulation frequency; when there are multiple candidate peak values, the frequencies are searched and selected in order from high to low according to the energy, and the frequency which forms the most prominent sideband pair with the synchronous frequency of the roller is selected as the modulation frequency; when it cannot be determined, the one with the maximum energy is taken; the positive sideband and the negative sideband are the synchronous frequency of the roller plus or minus the modulation frequency, respectively;

[0067] Step 23, calculate the ratio of the sum of the energy of the frequency components in the positive sideband and the negative sideband to the sum of the energy of the frequency components in the main frequency band, to obtain a modulation ratio, the main frequency band being a range of a preset percentage above and below the synchronous frequency of the roller; wherein the modulation ratio reflects the degree of energy distribution of the main frequency band to the sideband, and is used to reflect the modulation effect in the operation of the roller;

[0068] Step 24, calculate the ratio of the kurtosis of the main frequency band signal to the kurtosis of the signal in the full frequency range, to obtain a kurtosis ratio; wherein the kurtosis is the ratio of the fourth moment of the signal deviating from the mean value to the square of the variance;

[0069] Step 25, combine the modulation ratio and the kurtosis ratio to form a cycle and sideband fingerprint prototype vector. This vector not only contains information about energy distribution and modulation effect, but also can comprehensively represent the pulse characteristics of the signal.

[0070] In an embodiment of the present application, the generation of the contrast embedding vector and the time sequence feature vector includes:

[0071] Step 31, perform data enhancement processing on the frame feature set, the data enhancement including random movement of the time axis and band-limited noise mixing; specifically, the random movement of the time axis refers to a random shift of the entire feature sequence in the time dimension, which is used to eliminate the phase difference caused by different sampling starting points; the band-limited noise mixing refers to superimposing random noise in a predetermined frequency range to enhance the robustness of the model in a noisy environment. The frame feature set after data enhancement can more comprehensively cover the sound feature distribution under actual operating conditions;

[0072] Step 32, construct paired samples based on the enhanced frame feature set and the cycle and sideband fingerprint prototype vector, wherein the positive sample pair is the feature sequence of the same device under the same operating condition, and the negative sample pair is the feature sequence of different devices or different operating conditions, and the operating condition is determined according to the classification results of the load and the belt speed of the device; by this method, the consistency of the internal mechanism of the device and the difference across devices can be captured simultaneously in the contrast learning process;

[0073] Step 33, input the paired samples into an encoder and a projector, the encoder including a one-dimensional convolutional layer and a gated recurrent unit for extracting a time sequence feature vector, and the projector including a multi-layer fully connected network for compressing the time sequence feature vector to obtain a contrast embedding vector;

[0074] Step 34, construct a loss function and optimize the parameters of the encoder and the projector, the loss function including a contrast loss and a cycle regularization loss, and being combined by weighting with a preset weight, wherein the contrast loss is calculated based on the similarity of the contrast embedding vectors of the positive sample pair and the similarity of the contrast embedding vectors of the negative sample pair, and the cycle regularization loss is calculated based on the time sequence feature vectors before and after phase shifting of the same signal; specifically, the calculation formula of the loss function is: ;

[0075] wherein, represents the value of the loss function, and respectively represent the first weight coefficient and the second weight coefficient for balancing the contribution of the contrastive loss and the cycle regularization loss, P represents the positive sample pair, represents the set containing the corresponding positive sample j and all negative samples with sample i as the anchor, and respectively represent the contrastive embedding vectors of the i-th and j-th positive samples, represents the temperature coefficient for adjusting the distribution smoothness, i and j represent the index of the positive sample, and k represents the index of the negative sample, represents the contrastive embedding vector of the k-th negative sample, and N represents the number of samples, represents the cosine similarity function, represents and the cosine similarity of represents the m-th time sequence feature vector, represents the time sequence feature vector obtained by phase shifting the same signal, represents the square of the Euclidean distance of the result of ;

[0076] Step 35, the encoder and the projector processed in step 34 are used for forward calculation, and the contrastive embedding vector and the time sequence feature vector are output as the input of the time sequence modeling network.

[0077] Through the above optimization process, the optimized contrastive embedding vector and the time sequence feature vector can be obtained, and they are used as the input of the subsequent time sequence modeling network; by introducing the paired samples and the loss function, the consistency modeling of the embedding space to the equipment mechanism characteristics and the differentiation ability to the differences of different working conditions can be ensured at the same time; through the cycle regularization, the model is not sensitive to the phase change caused by the rotation of the roller, thereby enhancing the robustness under different working conditions.

[0078] In an embodiment of the present application, the data enhancement further comprises frequency band masking and time frame masking, wherein the frequency band masking refers to randomly selecting a continuous frequency band in the frequency dimension of the frame feature set and setting the features in the band to zero, and the frequency dimension of the frame feature set is composed of a plurality of frequency channels output by a mel filter bank, each channel corresponding to an energy component of a specific frequency range. By masking the continuous frequency channels, the loss of signals in some frequency bands can be simulated, so that the model does not rely on features of a specific frequency band during learning, thereby improving robustness in different noise environments. The time frame masking refers to randomly selecting a continuous frame in the time dimension of the frame feature set and setting the features in the frame to zero. The time dimension is composed of a frame sequence formed by dividing the running sound data according to a preset frame length and frame shift, and each frame corresponds to the sound features of a time segment. By masking the continuous time frames, the loss or interference of signals in some time segments can be simulated, so that the model can learn to maintain the recognition ability of the overall signal pattern under the condition of missing data during training.

[0079] By introducing frequency band masking and time frame masking, the model can avoid over-reliance on specific frequency features or local time features, thereby improving the robustness of the contrast learning network when facing signal loss, noise interference or working condition fluctuations. This masking mechanism is complementary to the aforementioned random time axis movement and band-limited noise mixing, and can improve the generalization ability and stability of the model in both frequency and time domains.

[0080] In an embodiment of the present application, the generation of the three types of probabilities comprises:

[0081] Step 41, segmenting the optimized time sequence feature vector, dividing the time sequence feature vector of continuous frames into segment-level feature sequences according to a preset length;

[0082] Step 42, inputting the segment-level feature sequences into a branch network for decomposition, wherein the low-frequency trend branch extracts long-term dependency features through a gated recurrent unit, the high-frequency impact branch extracts local impulse features through convolution operation and gating mechanism, and the long-term dependency features and the local impulse features are spliced in the same feature dimension to form an enhanced feature vector. Through this decomposition and fusion structure, both long-term and short-term features can be considered, and feature complementation and information enhancement can be realized in a unified feature space;

[0083] Step 43, inputting the enhanced feature vector into a classifier to output a classification raw output value vector containing normal, mild abnormality and severe abnormality, and performing normalization calculation through a softmax function to obtain three types of probabilities.

[0084] The embodiment can capture long-term evolution characteristics and instantaneous pulse characteristics at the same time through the synergistic effect of the low-frequency trend branch and the high-frequency impact branch, and realize fusion in the enhanced feature vector; through the classifier combined with softmax normalization, the enhanced features can be accurately mapped into three types of probabilities of normal, mild abnormality and severe abnormality, thereby providing a reliable basis for subsequent abnormal event identification and hierarchical processing.

[0085] It should be noted that the step 33 focuses on joint representation of the frame feature set and the periodic fingerprint prototype vector, specifically extracts a general time sequence feature vector through convolution and cyclic structure, and further compresses to form a contrast embedding vector in the projector. The core of this process is to establish a unified embedding representation space, so as to realize the discriminant ability across devices and working conditions through contrast loss. The step 42 performs branch decoupling processing based on the optimized time sequence feature vector; the low-frequency trend branch is used to extract long-term dependence features, and the high-frequency impact branch is used to capture local pulse features, and the two are spliced and fused in the same feature dimension to form an enhanced feature vector. The core of this process is feature enhancement for classification tasks, so that the model can utilize long-term trend information and short-term impact information at the same time.

[0086] In an embodiment of the present application, the calculation of the abnormal score and the determination of the abnormal event level include:

[0087] In step 51, a normal probability is subtracted as a classification contribution, a difference greater than zero in the sum of squares of the classification original output value vectors of adjacent two segments is taken as a spectral flux contribution, which is used to measure the cumulative change of abnormal energy of the signal in the time sequence evolution process, a modulation ratio is taken as a sideband contribution, which is used to reflect the abnormal enhancement of the signal spectrum sideband component relative to the main frequency band, and the three parts are weighted and summed according to the preset weight to obtain an abnormal score;

[0088] In step 52, the abnormal score sequence of the continuous segments is detected, the centralization abnormal score is calculated by subtracting the upper limit of the abnormal score, and the one-sided cumulative sum of the centralization abnormal score is calculated, when the one-sided cumulative sum exceeds the preset deviation threshold and lasts for more than three continuous segments, it is determined as an abnormal event; wherein the upper limit of the abnormal score is obtained by taking the median plus the quantile difference; the centralization abnormal score is obtained by comparing the abnormal score of each segment with the reference upper limit and calculating the difference between them; the one-sided cumulative sum means that when a segment of the centralization abnormal score is positive, it is added to the cumulative sum, and when it is negative, the cumulative sum is reset to zero; this method can amplify the abnormal trend deviating from the normal level continuously, without causing false judgment due to a single noise point;

[0089] In step 53, within the window corresponding to the abnormal event, the abnormal score is first subjected to exponential transformation, then the transformed score is averaged and taken as a natural logarithm to obtain an event intensity; specifically, the calculation formula of the event intensity is: wherein, represents the event intensity, Q represents the number of continuous segments contained in the abnormal time window, e represents the natural logarithm, represents the abnormal score of the qth segment in the abnormal time window; in this way, the numerical scale can be compressed while being amplified nonlinearly, so that the event intensity index is more suitable for comparison with the threshold value;

[0090] Step 54, determining as normal when the event intensity is lower than the first threshold value, determining as mild abnormal when the event intensity is between the first threshold value and the second threshold value, and determining as severe abnormal when the event intensity is not lower than the second threshold value.

[0091] The embodiment fuses the multi-dimension of classification contribution, spectrum flux contribution and sideband contribution, so that the abnormal score can not only reflect the deviation degree of the classification result, but also depict the signal energy change trend and spectrum modulation characteristics; the false positive rate can be effectively reduced through the joint constraint of cumulative deviation and threshold condition; the quantitative characterization and hierarchical identification of abnormal events can be realized through the nonlinear transformation and hierarchical determination of the event intensity, thereby significantly improving the accuracy and robustness of the belt conveyor abnormal detection and diagnosis.

[0092] In an embodiment of the present application, the step 6 specifically comprises:

[0093] Step 61, acquiring the abnormal event level, event intensity, device criticality and real-time load, and obtaining the priority score through weighted summation; wherein the device criticality represents the importance of the belt conveyor in the overall production system, which can be set according to the production link dependence and downtime loss degree; the real-time load refers to the running load level of the belt conveyor at the current time, reflecting its running intensity and potential risk;

[0094] Step 62, performing service level agreement mapping according to the priority score and the abnormal event level; specifically, through the preset service level agreement rules, different levels of abnormalities and priority scores are corresponded to the corresponding service timeliness and response requirements, so as to ensure that high-priority and high-risk events are quickly processed, while low-priority events are reasonably allocated resources, thereby improving the overall maintenance efficiency and resource utilization;

[0095] Step 63, under the premise of meeting the skill matching, the generated work order is dispatched to the maintenance personnel, the skill matching is judged based on the correspondence between the skill label required by the work order and the skill label of the maintenance personnel, and the shortest path distance between the work order location and the current location of the maintenance personnel is used as the dispatch priority condition. Wherein, the skill label is a feature label used to describe the professional ability of the maintenance personnel and the task demand of the work order; the skill label includes but is not limited to electrical maintenance, mechanical maintenance, vibration analysis, acoustic diagnosis and safety operation, etc.

[0096] The belt conveyor abnormal event disposal process is intelligently and efficiently realized by the dispatching strategy of comprehensively combining the priority score, the service level agreement mapping, and the skill matching and path optimization, which can not only guarantee the priority response of the key equipment abnormality, but also improve the rationality of the maintenance resource allocation, and significantly improve the response speed and processing efficiency of the overall operation and maintenance management of the belt conveyor.

[0097] In an embodiment of the present application, the treatment information includes treatment time, material consumption, replacement part code and treatment result; the treatment time is used to record the time consumption from the arrival of the maintenance personnel at the scene to the recovery of the equipment to operation, the material consumption is used to record the types and quantities of materials used in the maintenance process, the replacement part code is used to identify the specific part model replaced, and the treatment result is used to indicate whether the equipment recovers normal operation or whether there is residual hidden danger.

[0098] The equipment recovery operation time is determined according to the treatment time of the treatment information, and the downtime length is obtained in combination with the downtime start time;

[0099] The total cost includes the fixed cost corresponding to the work order type, the material cost and the downtime cost, and the downtime cost is obtained by multiplying the unit time downtime loss rate and the downtime length. In this way, the direct and indirect economic impact of the belt conveyor maintenance and disposal process on production and operation can be quantified.

[0100] It should be noted that the interval and the threshold size are set for ease of comparison, and the size of the threshold depends on the number of sample data and the base number set by the person skilled in the art for each group of sample data, as long as it does not affect the proportional relationship of the parameters and the quantized values. And the above formula is a dimensionless calculation of the value, and the formula is obtained by software simulation of a large amount of data to obtain a formula of the most recent real situation, and the preset parameters in the formula are set by the person skilled in the art according to the actual situation.

[0101] The embodiments of the present application are described above, but the present application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative and not limiting, and a person of ordinary skill in the art can make many forms under the inspiration of the present embodiment, which all belong to the protection of the present embodiment.

Claims

1. A method for managing auscultatory abnormalities in belt conveyors based on contrastive learning and temporal modeling, characterized in that, Includes the following steps: Step 1: Obtain the operating sound data and equipment parameters of the belt conveyor; extract frame features based on the operating sound data; and obtain the acoustic baseline under normal operating conditions. The equipment parameters include the idler diameter and belt speed. Step 2: Calculate the roller synchronization frequency based on the equipment parameters, and construct the period and side-band fingerprint prototype vector, where the period and side-band fingerprint prototype vector includes the modulation ratio and kurtosis ratio; Step 3: Input the frame features and periodicity with the sideband fingerprint prototype vector into the contrast learning network, perform contrast learning according to the set data augmentation and pairing rules, and generate contrast embedding vectors and temporal feature vectors; Step 4: Input the temporal feature vector into the temporal modeling network, and output the probabilities of three classes: normal, mildly abnormal, and severely abnormal, as well as the original classification output values. Step 5: Calculate the anomaly score based on the three probabilities, the original output value of the classification, the period and the side-band fingerprint prototype and the voiceprint baseline, generate anomaly events based on the anomaly scores, calculate the event intensity, and determine the anomaly event level according to the set threshold. Step 6: Calculate the priority score based on the abnormal event level, event intensity, equipment criticality, and real-time load, and generate and dispatch work orders according to the set service level agreement. Step 7: The maintenance personnel complete the on-site handling according to the work order and fill in the handling information. They calculate the total cost based on the handling information and downtime, and store the handling information and total cost in the database.

2. The belt conveyor auscultation anomaly management method based on contrastive learning and temporal modeling according to claim 1, characterized in that, Frame features are extracted from the operating sound data, and a voiceprint baseline is obtained under normal operating conditions, including: Step 11: Divide the running sound data into frames according to the first preset frame length and the second preset frame shift. Perform a fast Fourier transform on each frame and calculate the Mel frequency cepstral coefficients through the Mel filter bank. At the same time, extract the instantaneous energy in each frame to obtain the combination of Mel frequency cepstral coefficients and instantaneous energy. Step 12: Calculate the first-order difference and the second-order difference based on the Mel frequency cepstral coefficients, and combine the Mel frequency cepstral coefficients, the first-order difference, the second-order difference, and the instantaneous energy into a frame feature set; wherein, the first-order difference is obtained by the difference between adjacent frame features, and the second-order difference is obtained by the difference between adjacent first-order differences. Step 13: Under normal operating conditions, collect a set of frame features for multiple consecutive frames. The set of frame features includes energy values ​​for each frequency dimension obtained by the Mel filter bank. Calculate the median of the energy value sequence for each frequency dimension as the reference energy level for that frequency, and calculate the difference between the 95th percentile of the energy value sequence for that frequency and the median as the upper limit of the energy deviation for that frequency. Step 14: Form an ordered pair between the reference energy level and the upper limit of energy deviation for each frequency, and then form the acoustic baseline by combining all ordered pairs.

3. The belt conveyor auscultation anomaly management method based on contrastive learning and temporal modeling according to claim 1, characterized in that, Calculate the idler synchronization frequency based on equipment parameters, and construct a cycle and side-band fingerprint prototype, including: Step 21: Calculate the circumference based on the roller diameter and pi, and use the ratio of belt speed to circumference as the roller synchronization frequency. Step 22: Divide the key frequency bands with the idler synchronization frequency as the center, and select the frequency corresponding to the energy peak as the modulation frequency; the positive band and the negative band are the idler synchronization frequency plus or minus the modulation frequency, respectively. Step 23: Calculate the ratio of the sum of frequency component energy in the positive and negative band ranges to the sum of frequency component energy in the main frequency band range to obtain the modulation ratio. The main frequency band is the range of preset percentages above and below the idler roller synchronization frequency. Step 24: Calculate the ratio of the kurtosis of the main frequency band signal to the kurtosis of the full frequency range signal to obtain the kurtosis ratio; where kurtosis is the ratio of the fourth moment of the signal deviating from the mean to the square of the variance. Step 25: Combine the modulation ratio and kurtosis ratio to form the periodic and sideband fingerprint prototype vectors.

4. The belt conveyor auscultation anomaly management method based on contrastive learning and temporal modeling according to claim 1, characterized in that, The generation of embedding vectors and temporal feature vectors is compared, including: Step 31: Perform data augmentation processing on the frame feature set, the data augmentation including random time axis shifting and band-limited noise mixing; Step 32: Construct paired samples based on the enhanced frame feature set and the period and side-band fingerprint prototype vector. Positive sample pairs are feature sequences of the same device under the same operating conditions, while negative sample pairs are feature sequences of different devices or under different operating conditions. The operating conditions are determined based on the classification results of the device's load and belt speed. Step 33: Input the paired samples into the encoder and the projector. The encoder includes a one-dimensional convolutional layer and a gated recurrent unit for extracting temporal feature vectors. The projector includes a multi-layer fully connected network for compressing the temporal feature vectors to obtain contrast embedding vectors. Step 34: Construct a loss function and optimize the parameters of the encoder and projector. The loss function includes contrast loss and periodic regularization loss, which are weighted and combined by preset weights. The contrast loss is calculated based on the similarity between the contrast embedding vectors of positive sample pairs and the contrast embedding vectors of negative sample pairs. The periodic regularization loss is calculated based on the temporal feature vectors of the same signal before and after phase shift. Step 35: The encoder and projector processed in step 34 are used for forward computation to output the contrast embedding vector and temporal feature vector, which are used as inputs to the temporal modeling network.

5. The belt conveyor auscultation anomaly management method based on contrastive learning and temporal modeling according to claim 4, characterized in that, The data augmentation also includes frequency interval masking and time frame masking. Frequency interval masking refers to randomly selecting a continuous frequency interval in the frequency dimension of the frame feature set and setting the features of that interval to zero. Time frame masking refers to randomly selecting a continuous frame in the time dimension of the frame feature set and setting the features of that part to zero.

6. The belt conveyor auscultation anomaly management method based on contrastive learning and temporal modeling according to claim 1, characterized in that, The generation of the three types of probabilities includes: Step 41: The optimized temporal feature vector is segmented, and the temporal feature vectors of consecutive frames are divided into segment-level feature sequences according to a preset length. Step 42: Input the segment-level feature sequence into the branch network for decomposition. The low-frequency trend branch extracts long-term dependency features through gated recurrent units, and the high-frequency impulse branch extracts local impulse features through convolution operations and gating mechanisms. The long-term dependency features and local impulse features are concatenated on the same feature dimension to form an enhanced feature vector. Step 43: Input the enhanced feature vector into the classifier, and output the original output value vector containing normal, mildly abnormal and severely abnormal classification values. Then, normalize the vector using the softmax function to obtain the probabilities of the three classes.

7. The belt conveyor auscultation anomaly management method based on contrastive learning and temporal modeling according to claim 1, characterized in that, The calculation of anomaly scores and the determination of anomaly event levels include: Step 51: Subtract the normal probability from 1 as the classification contribution, take the part of the difference between the squares of the original output value vectors of two adjacent segments that is greater than zero as the spectral flux contribution, take the modulation ratio as the sideband contribution, and sum the three parts according to the preset weights to obtain the anomaly score. Step 52: Detect abnormal score sequences based on continuous segments. Calculate the centered abnormal score by subtracting it from the upper limit of the abnormal score, and perform a one-sided cumulative sum on the centered abnormal score. When the one-sided cumulative sum exceeds the preset deviation threshold and continues for more than three consecutive segments, it is determined to be an abnormal event. Step 53: In the window corresponding to the abnormal event, first perform an exponential transformation on the abnormal score, then take the mean and natural logarithm of the transformed score to obtain the event intensity. Step 54: When the event intensity is below the first threshold, it is determined to be normal; when the event intensity is between the first threshold and the second threshold, it is determined to be slightly abnormal; when the event intensity is not lower than the second threshold, it is determined to be severely abnormal.

8. The belt conveyor auscultation anomaly management method based on contrastive learning and temporal modeling according to claim 1, characterized in that, Step 6 specifically includes: Step 61: Obtain the abnormal event level, event intensity, device criticality, and real-time load, and obtain the priority score by weighted summation; Step 62: Perform service level agreement mapping based on priority scores and abnormal event levels; Step 63: Under the premise of satisfying skill matching, the generated work order is dispatched to the maintenance personnel. The skill matching is based on the correspondence between the skill tags required by the work order and the skill tags of the maintenance personnel, and the shortest path distance between the work order location and the current location of the maintenance personnel is used as the dispatch priority condition.

9. The belt conveyor auscultation anomaly management method based on contrastive learning and temporal modeling according to claim 1, characterized in that, The disposal information includes disposal time, material consumption, replacement part codes, and disposal results; The equipment recovery time is determined based on the processing time of the handling information, and the downtime is obtained by combining the downtime start time; The total cost includes: fixed costs, material costs, and downtime costs corresponding to the work order type. Downtime costs are obtained by multiplying the downtime loss rate per unit time by the downtime duration.

Images