Multi-source data cooperative monitoring method for faults of transmission drum of belt conveyor

Abstract

The invention provides a multi-source data cooperative monitoring method for faults of a transmission drum of a belt conveyor, relates to the field of state monitoring and fault diagnosis of mechanical equipment, and solves the problem of automatic monitoring and analysis of the transmission drum. The method comprises the following steps: collecting multi-source data required by transmission roller fault analysis and transmitting the multi-source data in real time; judging whether the analysis algorithm needs to be subjected to first training and incremental training or not, if any training needs to be carried out, entering an algorithm training step, and otherwise, skipping to a cooperative monitoring step; the analysis algorithm is trained and updated by using multi-source data, and the updated analysis algorithm is transmitted to the cooperative monitoring step; rotating speed analysis, visual analysis and time sequence analysis are executed in a parallel mode, analysis results are output, real-time judgment and prediction results of faults of the transmission roller are obtained, and data are collected again for the next monitoring cycle; according to the invention, real-time and omnibearing monitoring of the running state of the transmission roller is realized, and the overall running efficiency and safety of the belt conveyor are effectively improved.

Description

Technical Field

[0001] The invention relates to the technical field of mechanical equipment state monitoring and fault diagnosis, and in particular to a multi-source data collaborative monitoring method for belt conveyor transmission roller faults. Background Art

[0002] Belt conveyors are one of the most important material transportation equipment at present, which can realize continuous material transportation. With the continuous advancement and development of technology, a variety of new belt conveyors such as tubular belt conveyors, wheel-rail belt conveyors, and cable belt conveyors have been developed and put into production to meet the material transportation needs in different environments. These new conveyors not only improve the efficiency and flexibility of material transportation, but also put forward higher requirements on the stability and safety of the equipment.

[0003] As the core driving component of the belt conveyor, the drive roller plays a vital role in various types of belt conveyors. It is usually driven by a motor or a reducer, and the power is transmitted through the friction between the conveyor belt and the conveyor belt. The state of the drive roller is directly related to the overall performance of the conveying system. Once damaged or malfunctions, it will not only lead to a significant decrease in conveying efficiency, but also may cause damage to the belt and even cause serious consequences such as casualties. Therefore, ensuring the stable and reliable operation of the drive roller is crucial to ensuring production safety and improving production efficiency.

[0004] Traditionally, the operation monitoring of belt conveyors mainly relies on two methods: physical sensors and manual regular inspections. However, this monitoring method has obvious limitations when facing harsh working environments. On the one hand, the frequency of manual monitoring is limited, and it is difficult to ensure the accuracy of the monitoring results due to the professional level and experience of the staff; on the other hand, although physical sensors can make up for the shortcomings of manual monitoring to a certain extent, their monitoring range and accuracy are also limited, especially under complex and changeable working conditions. It is difficult for a single sensor to fully capture all abnormal information of the transmission roller. In addition, traditional monitoring methods are also accompanied by high labor costs and low fault response speeds, which not only increases maintenance costs, but may also lead to failures that cannot be discovered and handled in a timely manner, thereby affecting the long-term stable operation of the equipment.

[0005] In view of the above situation, developing an efficient and accurate transmission roller automatic monitoring technology has become an urgent problem to be solved. Summary of the invention

[0006] The purpose of the present invention is to systematically solve the problem of automated monitoring and analysis of drive rollers, and therefore a multi-source data collaborative monitoring method for belt conveyor drive roller faults is proposed; the present invention realizes real-time and all-round monitoring of the operating status of the drive roller, which can not only improve the accuracy and timeliness of fault detection, but also effectively reduce maintenance costs and improve the overall operating efficiency and safety of the belt conveyor.

[0007] The present invention adopts the following technical solutions to achieve the purpose:

[0008] A multi-source data collaborative monitoring method for belt conveyor transmission roller faults, the method comprising the following steps:

[0009] S1. Data collection and transmission: Collect multi-source data required for transmission roller fault analysis and transmit it in real time;

[0010] S2, training condition judgment: judge whether the analysis algorithm needs to be trained for the first time and incremental training in turn; if any training is required, enter the algorithm training step; if no training is required, jump to the collaborative monitoring step;

[0011] S3, algorithm training: use multi-source data to train and update the analysis algorithm, and after the training is completed, transmit the updated analysis algorithm to the collaborative monitoring step;

[0012] S4, collaborative monitoring: perform speed analysis, visual analysis and time series analysis in parallel and output their respective analysis results;

[0013] S5. Result output: Based on the analysis results, the real-time judgment result and prediction result of the transmission roller fault are output, and at the same time, the data collection and transmission step is returned to carry out the next round of monitoring cycle.

[0014] Specifically, in the training condition judgment step of S2, first determine whether the analysis algorithm needs to be trained for the first time; if the first training is required, set time t=1, jump to the data collection and transmission step and set time t=t+1, until time t meets the required number of training times, enter the algorithm training step; if the first training is not required, continue to determine whether incremental training is required; if incremental training is required, enter the algorithm training step; if incremental training is not required, jump to the collaborative monitoring step;

[0015] In the result output step of S5, the transmission roller fault includes bearing fault and rubber lagging fault; after respectively outputting the real-time judgment results and prediction results corresponding to the two types of faults, let time t=t+1 and return to the data collection and transmission step to carry out the next round of monitoring cycle.

[0016] Preferably, in the algorithm training step of S3, if this is the first training, then after the multi-source data collection is completed, the data set formed by the multi-source data is used to train the visual analysis algorithm and the time series analysis algorithm; if this is incremental training, then after the multi-source data collection is completed, only the data set formed by the multi-source data is used to train the time series analysis algorithm.

[0017] Preferably, in a monitoring cycle, incremental training and collaborative monitoring are performed independently, that is, collaborative monitoring is performed while incremental training is performed; the duration of an incremental training is one or more monitoring cycle rounds, and the analysis algorithm currently used in the collaborative monitoring step is not updated before the completion of an incremental training; when the incremental training is completed, the analysis algorithm is updated and used in the collaborative monitoring step of the next monitoring cycle round.

[0018] Furthermore, the training of the visual analysis algorithm includes a first training data set construction process and a first parameter solution process, as follows:

[0019] In the process of constructing the first training data set, for any time t, the collected image data is recorded as in i=1,2,…,k represent the image data collected under k different camera positions and shooting angles respectively; Respectively represent the damaged area of ​​the belt conveyor surface rubber and the material coverage area; at any time t, the first training sample is represented as As of time t, the first training data set is represented as

[0020] In the first parameter solving process, for the visual analysis algorithm Ψ, the loss function Loss1 is set, and the optimization goal is to minimize Loss1, and the parameters of the visual analysis algorithm Ψ are solved using the first training data set; the model corresponding to the visual analysis algorithm Ψ includes a first input layer, a shared layer, a specific task layer and a first output layer; wherein the first input layer is based on the training sample Input F i , i=1,2,…,t as input; the shared layer is used to extract the shared features of the input data between different tasks; the specific task layer consists of the rubber damage analysis and the material accumulation analysis; the first output layer is used to output the corresponding input F i ,i=1,2,…,t damaged lagging area And the area of ​​material accumulation on the drum surface and Calculated or estimated by the number of pixels covering the damaged lagging and the accumulated material; the loss function Loss1 is used to measure the true value and predicted values The gap between.

[0021] Furthermore, the training of the time series analysis algorithm includes a second training data set construction process and a second parameter solution process, as follows:

[0022] In the process of constructing the second training data set, for any time t, the collected time series data is recorded as S t =[A t ,N t ,E t ,I t ], where A t Indicates vibration data, N t represents noise data, E t Indicates temperature data, I t represents the current data; as the belt conveyor runs, the data set formed after data collection is represented as {S1, S2, …, S t}, let the linear sliding window size be w, then the input at time t-1 is represented as {S t-1 ,S t-2 ,…,S t-w}; The second training sample corresponding to time t is expressed as Then, as of time t, the second training data set is represented as

[0023] In the second parameter solving process, for the time series analysis algorithm Φ, the loss function Loss2 is set, and the optimization goal is to minimize Loss2. The parameters of the time series analysis algorithm Φ are solved using the second training data set. The model corresponding to the time series analysis algorithm Φ includes a second input layer, a feature extraction layer, a fusion layer, and a second output layer. The second input layer is based on the training samples. Input {S i-1 ,S i-2 ,…,S i-w}, i = w, w + 1, ..., t as input; the feature extraction layer includes vibration feature extraction, noise feature extraction, temperature feature extraction and current feature extraction, which are used to extract features of different types of data respectively; the fusion layer is used to splice the four features; the second output layer is used to output the corresponding input {S i-1 ,S i-2 ,…,S i-w},i=w,w+1,…,t’s predicted value, which can be expressed as in Each parameter is the predicted value corresponding to the collected time series data; the loss function Loss2 is used to measure the true value S i and predicted values The gap between.

[0024] Further, in the collaborative monitoring step of S4, for each time t, the speed analysis, visual analysis and time series analysis are performed in parallel, and the respective analysis results are output;

[0025] In the speed analysis, the speed of the driving drum and the driven drum is analyzed to determine whether slippage occurs, and then determine whether the driving drum is in a normal state, a rubber lagging damage state, or a bearing damage state;

[0026] In visual analysis, the image of the lagging surface is analyzed to determine whether there is material accumulation and whether the lagging is damaged, and then determine whether the drive pulley is in a normal state or a damaged state, and at the same time predict whether the lagging will be damaged;

[0027] In time series analysis, vibration data, noise data, temperature data and current data are analyzed collaboratively to perform single-step judgment and future trend analysis, and output whether a fault has occurred and a prediction of future fault trends.

[0028] Specifically, speed analysis includes bearing fault monitoring and lagging fault monitoring;

[0029] In bearing fault monitoring: collect the speed data of the transmission roller shaft and the driven roller shaft, set the rated speed and the transmission roller speed fluctuation range; determine the transmission roller speed based on the collected data and the set data; when the speed is determined to be lower than the preset threshold, output the judgment result of "bearing damage"; when the speed is determined to be within the preset normal range, perform bearing trend analysis;

[0030] In the bearing trend analysis, by presetting the abnormal trend threshold, it is determined whether the transmission roller has a trend of increasing rotation resistance; if so, it is predicted that "the bearing may fail", the prediction result is output and the staff is reminded to check the bearing failure;

[0031] In the monitoring of rubber lagging faults: set the speed difference fluctuation range and calculate it. If the calculation result is greater than the preset threshold, it is determined that the speed of the driven roller is lower than the speed of the driving roller, indicating that there is a slip problem, and predict that "the rubber lagging may be damaged". The prediction result is output and the staff is reminded to troubleshoot the rubber lagging fault; if the calculation result is within the preset normal range, the rubber lagging trend analysis is performed;

[0032] In the rubber lagging trend analysis, by presetting the abnormal trend threshold, it is determined whether there is a slipping trend; if so, the prediction result of "slipping may occur" is output and the staff is reminded to check the fault of rubber lagging damage on the surface of the transmission roller; otherwise, the monitoring result of "no obvious abnormality" is output.

[0033] Specifically, the visual analysis is based on the collected data, and the visual analysis algorithm is used to perform the rubber lagging damage analysis and material accumulation analysis, and output the rubber lagging damage area and the material accumulation area on the surface of the transmission drum;

[0034] The specific process of rubber lagging damage analysis is as follows: set the rubber lagging damage judgment threshold, calculate it in combination with the surface area of ​​the transmission roller, and if the calculation result does not meet the rubber lagging damage judgment threshold range, output the judgment result of "rubber lagging damage"; otherwise, perform rubber lagging trend analysis;

[0035] In the rubber lagging trend analysis, by presetting the abnormal trend threshold, it is judged whether there is a trend of rubber lagging damage or increasing shedding area; if so, it is predicted that "the rubber lagging surface may be damaged", the prediction result is output and the staff is reminded to check the fault of rubber lagging damage on the surface of the transmission roller;

[0036] The specific process of material accumulation analysis is as follows: set the material accumulation judgment threshold, calculate and determine whether the area of ​​material accumulation on the surface of the transmission roller is too large, and if so, output the prediction result and remind the staff to deal with it; otherwise, perform material trend analysis, and determine whether there is a trend of material accumulation by presetting the abnormal trend threshold, and output the corresponding prediction result.

[0037] Specifically, the specific process of time series analysis is as follows: based on the collected data, the characteristics of vibration data, noise data, temperature data and current data are extracted using the time series analysis algorithm, and after calculation and processing, the predicted values ​​corresponding to vibration, noise, temperature and current are output; the abnormal judgment thresholds of vibration, noise, temperature and current are set respectively, and the calculated predicted values ​​are compared with the four types of abnormal judgment thresholds; if any of the predicted values ​​does not meet the abnormal judgment threshold of the corresponding category, it is determined that the transmission roller bearing is running abnormally, and the judgment result of "bearing fault" is output; otherwise, abnormal trend analysis is performed;

[0038] In the abnormal trend analysis, abnormal trend thresholds are set for vibration, noise, temperature and current respectively. After calculating the abnormal trends corresponding to each category and forming a judgment vector, they are compared with the reference vector formed by the set abnormal trend threshold. If the judgment vector of any category does not meet any threshold element in its corresponding reference vector, it is judged that there is a trend of increasing values ​​in the corresponding category, and it is predicted that "the bearing operation may be abnormal". The prediction result is output and the staff is reminded to check the bearing fault; otherwise, the monitoring result of "no obvious abnormality" is output.

[0039] In summary, due to the adoption of this technical solution, the beneficial effects of the present invention are as follows:

[0040] The method of the present invention is particularly aimed at health monitoring of the transmission roller of the belt conveyor. The method incorporates two common faults of the transmission roller, namely, rubber damage fault and bearing damage fault, into the monitoring and analysis at the same time, and provides more comprehensive technical support compared with the existing monitoring technology.

[0041] The method of the present invention not only focuses on the operating data of the transmission roller itself, but also incorporates the data of other key components (such as driven roller data, material data, and motor data) into the same monitoring system for collaborative analysis. This comprehensive application of multi-source data makes the analysis results more accurate and comprehensive, and can more effectively identify and diagnose potential faults of the transmission roller.

[0042] In addition, the method of the present invention not only considers the signs of fault occurrence during fault monitoring, but also covers the specific manifestations of fault occurrence. This method can more comprehensively reflect the operating status of the transmission roller under different working conditions, which is helpful for early detection of fault signs and timely taking preventive measures to avoid further deterioration of the fault.

[0043] Through the collaboration of multi-source data, the method of the present invention provides specific ways of trend prediction and fault diagnosis. These means not only improve the accuracy and timeliness of fault prediction, but also provide reliable support for fault monitoring and management. Specifically, the trend prediction function can help users understand the health status of the transmission roller in advance and formulate a reasonable maintenance plan; while the fault diagnosis function can quickly locate the problem when a fault occurs, guide maintenance personnel to take quick action, and reduce downtime and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the overall process of the method of the present invention;

[0045] Figure 2 The following is a schematic diagram of the classification and causes of belt conveyor drive roller failures;

[0046] Figure 3 It is a schematic diagram of the flow of rotation speed analysis in the method of the present invention;

[0047] Figure 4 Schematic diagram of the process of visual analysis in the method of the present invention;

[0048] Figure 5 Schematic diagram of the process of time series analysis in the method of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0050] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. 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.

[0051] Example 1

[0052] A multi-source data collaborative monitoring method for belt conveyor transmission roller faults. Figure 1 The overall process of the method is shown and can be viewed simultaneously; the specific steps of the method include: data collection and transmission, training condition judgment, algorithm training, collaborative monitoring, and result output. Figure 2 The application object of the method is shown, that is, the classification and causes of belt conveyor drive roller failures, etc. The embodiment then introduces the details of each step in the method in detail.

[0053] First, we can make a general introduction. The method includes the following steps:

[0054] S1. Data collection and transmission: Collect multi-source data required for transmission roller fault analysis and transmit it in real time;

[0055] S2, training condition judgment: judge whether the analysis algorithm needs to be trained for the first time and incremental training in turn; if any training is required, enter the algorithm training step; if no training is required, jump to the collaborative monitoring step;

[0056] S3, algorithm training: use multi-source data to train and update the analysis algorithm, and after the training is completed, transmit the updated analysis algorithm to the collaborative monitoring step;

[0057] S4, collaborative monitoring: perform speed analysis, visual analysis and time series analysis in parallel and output their respective analysis results;

[0058] S5. Result output: Based on the analysis results, the real-time judgment result and prediction result of the transmission roller fault are output, and at the same time, the data collection and transmission step is returned to carry out the next round of monitoring cycle.

[0059] In the data collection and transmission step, the collected multi-source data can be directly transmitted in real time to the training condition judgment step to perform related operations. The multi-source data here include: driving roller speed, driven roller speed, roller surface image, vibration data, noise data, temperature data, and current data.

[0060] Among them, the speed of the driving roller and the driven roller is obtained by the magnetic speed sensor arranged on the transmission; the image of the roller surface is taken by the high-definition camera arranged on the outside of the drive frame; the vibration data and temperature data are obtained by the temperature vibration sensor arranged at the bearing end of the driving roller; the noise data is obtained by the noise sensor arranged on the bearing side; the current data is the motor current.

[0061] In the training condition judgment step, determine whether the analysis algorithm needs to be trained for the first time. If so, set time t=1, jump to the data collection and transmission step and set t=t+1, until t=1000 and then enter the algorithm training step (in this embodiment, t is 1 second, and the belt conveyor is idling when collecting data); if not, determine whether incremental training is required (in this embodiment, the training is restarted every 2 hours). If so, enter the algorithm training step; if not, jump directly to the collaborative monitoring step.

[0062] In the algorithm training step, it is necessary to determine whether it is the first training. If so, after the data collection is completed, the collected data sets are used to train the visual analysis algorithm and the time series analysis algorithm. If not, it indicates incremental training, skipping the training of the visual analysis algorithm and only training the time series analysis algorithm. After the training is completed, the updated analysis algorithm is transmitted to the collaborative monitoring step.

[0063] It should be noted here that the incremental training time may be long and last for multiple monitoring cycles. Therefore, before the incremental training is completed, no changes are made to the original algorithm in the collaborative monitoring step, and the analysis algorithm before the update is still used for collaborative monitoring, that is, incremental training and collaborative monitoring are performed independently at the same time. For example, a monitoring cycle takes 30 seconds, and incremental training takes 10 minutes; then in these 20 monitoring cycles, the parameters of the analysis algorithm are not updated.

[0064] In this embodiment, the training of the visual analysis algorithm includes the following steps:

[0065] 1. Construction of the first training data set: For any time t, the collected image data can be expressed as in They represent the image data collected by camera 1 viewing angle 1, camera 2 viewing angle 2, camera 3 viewing angle 3, and camera 4 viewing angle 4, respectively. They represent the damaged area of ​​the belt conveyor surface rubber and the material coverage area. The training data is collected when the conveyor belt is idling. Then at any time t, the first training sample can be expressed as As of time t, the first training data set can be expressed as

[0066] 2. First parameter solution: For the visual analysis algorithm Ψ, set the loss function Loss1 to As the first training data set, the optimization goal is to minimize Loss1 and solve the parameters of the visual analysis algorithm Ψ.

[0067] In this embodiment, the model corresponding to the visual analysis algorithm Ψ includes a first input layer, a shared layer, a specific task layer, and a first output layer. The first input layer is based on the first training sample Input F i ,i=1,2,…,t as input; the shared layer is used to extract the shared features of the input data between different tasks. It consists of two two-dimensional convolutional layers and one Flatten layer. The activation function of the two-dimensional convolutional layer is the ReLU function; the specific task layer consists of rubber damage analysis and material accumulation analysis. The two specific tasks are respectively composed of a fully connected layer, a dropout layer, and a fully connected layer. The activation functions of the two fully connected layers are ReLU function and Sigmoid function respectively; the first output layer is used to output the corresponding input F i ,i=1,2,…,t damaged lagging area And the area of ​​material accumulation on the drum surface and Calculated or estimated by the number of pixels covering damaged lagging and accumulated material.

[0068] In this embodiment, the loss function Loss1 of the visual analysis algorithm Ψ is set to a binary cross entropy function to measure the true value and predicted values The gap between them; stochastic gradient descent is used to solve its algorithm parameters.

[0069] In this embodiment, the training of the time series analysis algorithm includes the following steps:

[0070] 1. Construction of the second training data set: For any time t, the collected time series data can be expressed as S t =[A t ,N t ,E t ,I t ], where A t Indicates vibration data, N t represents noise data, E t Indicates temperature data, I t Indicates current data.

[0071] As the belt conveyor runs, the collected data set can be expressed as {S1, S2, …, S t}. Let the linear sliding window size be w, then the input at time t-1 can be expressed as {S t-1 ,S t-2 ,…,St-w}. The second training sample at time t can be expressed as As of time t, the second training data set can be expressed as

[0072] 2. Second parameter solution: For the time series analysis algorithm Φ, set the loss function Loss2 to As the second training data set, the optimization goal is to minimize Loss2 and solve the parameters of the time series analysis algorithm Φ.

[0073] In this embodiment, the model corresponding to the time series analysis algorithm Φ includes a second input layer, a feature extraction layer, a fusion layer, and a second output layer. The second input layer is based on the second training sample Input {S i-1 ,S i-2 ,…,S i-w},i=w,w+1,…,t as input; the feature extraction layer includes vibration feature extraction, noise feature extraction, temperature feature extraction, and current feature extraction. Among them, vibration feature extraction and noise feature extraction are respectively composed of two one-dimensional convolution layers and one LSTM layer, and the activation function of the convolution layer is the ReLU function; temperature feature extraction and current feature extraction are respectively composed of one LSTM layer; the fusion layer is a Concatenate layer, which is used to splice the four features. The second output layer is composed of two connected fully connected layers, and the output value of the output layer is the corresponding input {S i-1 ,S i-2 ,…,S i-w},i=w,w+1,…,t’s predicted value can be specifically expressed as in

[0074] In this embodiment, the activation functions of the time series analysis algorithm Φ are composed of ReLU function and Sigmoid function respectively; the loss function Loss2 is set to mean square error, which is used to measure the true value S i and predicted values The gap between them; use the Adam algorithm to solve the parameters of the time series analysis algorithm Φ.

[0075] In the collaborative monitoring step, for each time t, the speed analysis, visual analysis and time series analysis are performed in parallel, and their respective analysis results are output, where:

[0076] Speed ​​analysis is used to analyze the speed of the drive roller and the driven roller to determine whether slippage occurs, whether the drive roller is in a normal state, a rubber lagging damage state, or a bearing damage state;

[0077] Visual analysis is used to analyze the image of the lagging surface to determine whether there is material accumulation and whether the lagging is damaged, whether the pulley is in a normal state, the lagging is damaged, and to predict whether the lagging will be damaged;

[0078] Time series analysis collaboratively analyzes vibration data, noise data, temperature data, and current data to perform single-step judgment and future trend analysis, thereby outputting whether a fault has occurred and a prediction of future fault trends.

[0079] In the result output step, the real-time judgment result and prediction result of the bearing fault and the real-time judgment result and prediction result of the rubber lagging fault are output, and after the time t=t+1, the data collection step is returned to perform the next cycle.

[0080] In summary, the multi-source data collaborative monitoring method of the present invention not only improves the accuracy and reliability of transmission roller fault monitoring, but also significantly enhances the overall operating efficiency and safety of the belt conveyor, providing a strong technical guarantee for industrial production.

[0081] Example 2

[0082] Based on Example 1, this example provides a detailed process introduction for the rotation speed analysis in the collaborative monitoring step. Figure 3 The process diagram is given.

[0083] For each time t, the speed data collected by the speed sensor can be expressed as in and Respectively represent the speeds collected by the conventional roller shaft and the driven roller shaft. The rated speed is set to v * =60r / min, set the transmission roller speed fluctuation range Let t denote the current time.

[0084] calculate judge Is it greater than γ? v If yes, it is determined that the transmission roller speed is significantly lower than the normal value, indicating that the running resistance of the transmission roller has increased significantly, and there is a problem of rotation resistance, and the judgment result of "bearing damage" is output; if no, perform trend analysis according to the following steps:

[0085] 1. Let the number of analysis steps be p = 600 and the abnormal trend threshold be

[0086] 2. Calculate the abnormal trend value in is an indicator function. When x is true, otherwise

[0087] 3. Judgment Is it greater than If so, it is judged that there is a trend of increasing rotational resistance, and it is predicted that "the bearing may fail". The prediction result is output and the staff is reminded to check the bearing failure.

[0088] At the same time, this embodiment also sets the speed difference fluctuation range value The speed difference is the difference between the driving drum speed and the driven drum speed.

[0089] Further calculation judge Is it greater than If yes, it is determined that the speed of the driven roller is significantly lower than that of the driving roller, indicating a slip problem. It is predicted that "the rubber lagging may be damaged", the analysis results are output, and the staff is reminded to check other slip reasons such as insufficient tension and belt surface damage; if no, perform trend analysis as follows:

[0090] 1. Let the number of analysis steps be p = 600 and the abnormal trend threshold be

[0091] 2. Calculate abnormal trends in is an indicator function. When x is true, otherwise

[0092] 3. Judgment Is it greater than If so, it is judged that there is a tendency to slip, and the prediction result of "slipping may occur" is output to remind the staff to check the fault of damage to the rubber coating on the surface of the transmission roller; if not, the monitoring result of "no obvious abnormality" is output.

[0093] Example 3

[0094] Based on Example 1, this example provides a detailed process introduction for the visual analysis in the collaborative monitoring step. Figure 4 The process diagram is given.

[0095] First, the training data set Or data input F t The shared layer transfers the extracted features to the specific task layer. After the specific task layer performs the analysis of the damage of the rubber and the accumulation of materials, the information is transferred to the output layer for the output of the damage area of ​​the rubber and the accumulation area of ​​materials. For any time t, the output of the output layer is in The area of ​​the damaged lagging and The surface area of ​​the drum is represented by a. Set the threshold for judging the damage of the rubber lagging Let t denote the current time.

[0096] calculate judge Is it greater than If yes, it is determined that the damaged area of ​​the rubber lagging is too large, indicating that the rubber lagging of the transmission pulley has obviously fallen off, and the judgment result of "rubber lagging damage" is output; if no, perform trend analysis according to the following steps:

[0097] 1. Let the number of analysis steps be p = 600 and the abnormal trend threshold be

[0098] 2. Calculate abnormal trends in is an indicator function. When x is true, otherwise

[0099] 3. Judgment Is it greater than If so, it is judged that there is a trend of damage to the rubber bag or an increase in the area of ​​detachment, and it is predicted that "the surface of the rubber bag may be damaged". The prediction result is output and the staff is reminded to check for abnormalities.

[0100] At the same time, this embodiment further sets a material accumulation judgment threshold calculate judge Is it greater than If yes, it is determined that too much material is accumulated on the surface of the drum, and it is predicted that "the drum rubber may be damaged due to material getting stuck and squeezed by the conveyor belt". The analysis results are output to remind the staff to clean the drum surface and check whether the cleaner is ineffective; if not, perform trend analysis according to the following steps:

[0101] 1. Let the number of analysis steps be p = 600 and the abnormal trend threshold be

[0102] 2. Calculate abnormal trends in is an indicator function. When x is true, otherwise

[0103] 3. Judgment Is it greater than If so, it is judged that there is a tendency of material accumulation, and the prediction result of "material accumulation may occur and damage the roller lagging" is output, and the staff is reminded to pay attention to cleaning the roller surface and check whether the cleaner fails; if not, the monitoring result of "no obvious abnormality" is output.

[0104] Example 4

[0105] Based on Example 1, this example provides a detailed process introduction for the time series analysis in the collaborative monitoring step. Figure 5 The process diagram is given.

[0106] First, the training data set Or data input {S t-1 ,S t-2 ,…,S t-w} is passed to the feature extraction layer and fusion layer for training algorithms or analyzing results. The feature extraction layer extracts the features of vibration data, noise data, temperature data, and current data respectively and transmits the information to the fusion layer. The fusion layer fuses all feature information together and passes the fusion result to the output layer. For any time t, the output of the output layer can be expressed as in Respectively represent the predicted values ​​of vibration, noise, temperature, and current at time t. Set the abnormal judgment thresholds of vibration, noise, temperature, and current respectively

[0107] The following parameters are calculated:

[0108]

[0109] For vibration, noise, temperature, and current, judge Is it greater than If any one of the items is true, the transmission roller bearing is judged to be running abnormally, and the judgment result of "bearing fault" is output; if not, the trend analysis is performed according to the following steps:

[0110] 1. Let the number of analysis steps be p = 600 and the abnormal trend threshold be

[0111] 2. Calculate the following abnormal trend parameters:

[0112]

[0113] in

[0114]

[0115] in

[0116]

[0117] in

[0118]

[0119] in

[0120] is an indicator function. When x is true, otherwise

[0121] 3. For vibration, noise, temperature, and current, determine the vector Is each element in greater than If any one of the items is true, it can be judged that the corresponding vibration, noise, temperature, and current have a trend of increasing values, predicting that "bearing operation may be abnormal", output the prediction result and remind the staff to check the transmission roller bearing; if not, the monitoring result of "no obvious abnormality" is output.

[0122] The methods in the above embodiments can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, they can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0123] Computer instructions may be stored in or transmitted via a computer-readable storage medium. A computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. Available media may be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid state disks (SSDs), etc.).

Claims

1. A multi-source data collaborative monitoring method for belt conveyor drive roller faults, characterized in that: The method comprises the following steps: S1. Data collection and transmission: Collect multi-source data required for transmission roller fault analysis and transmit it in real time; S2, training condition judgment: judge whether the analysis algorithm needs to be trained for the first time and incremental training in turn; if any training needs to be performed, enter the algorithm training step; If no training is required, jump to the collaborative monitoring step; S3, algorithm training: use multi-source data to train and update the analysis algorithm, and after the training is completed, transmit the updated analysis algorithm to the collaborative monitoring step; S4, collaborative monitoring: perform speed analysis, visual analysis and time series analysis in parallel and output their respective analysis results; S5. Result output: Based on the analysis results, the real-time judgment result and prediction result of the transmission roller fault are output, and at the same time, the data collection and transmission step is returned to carry out the next round of monitoring cycle.

2. The multi-source data collaborative monitoring method according to claim 1, characterized in that: In the training condition judgment step of S2, first determine whether the analysis algorithm needs to be trained for the first time; if it needs to be trained for the first time, set time t=1, jump to the data collection and transmission step and set time t=t+1, until time t meets the required number of training times, then enter the algorithm training step; If the first training is not required, continue to determine whether incremental training is required; If incremental training is required, enter the algorithm training step; if incremental training is not required, jump to the collaborative monitoring step; In the result output step of S5, the transmission roller fault includes bearing fault and rubber lagging fault; after respectively outputting the real-time judgment results and prediction results corresponding to the two types of faults, let time t=t+1 and return to the data collection and transmission step to carry out the next round of monitoring cycle.

3. The multi-source data collaborative monitoring method according to claim 1, characterized in that: In the algorithm training step of S3, if this is the first training, after the multi-source data collection is completed, the data set formed by the multi-source data is used to train the visual analysis algorithm and the time series analysis algorithm; if this is incremental training, after the multi-source data collection is completed, only the data set formed by the multi-source data is used to train the time series analysis algorithm.

4. The multi-source data collaborative monitoring method according to claim 3 is characterized in that: In a monitoring cycle, incremental training and collaborative monitoring are performed independently, that is, collaborative monitoring is performed while incremental training is performed; an incremental training lasts for one or more monitoring cycles, and the analysis algorithm currently used in the collaborative monitoring step is not updated before an incremental training is completed; when the incremental training is completed, the analysis algorithm is updated and used in the collaborative monitoring step of the next monitoring cycle.

5. The multi-source data collaborative monitoring method according to claim 3, characterized in that: The training of the visual analysis algorithm includes the first training data set construction process and the first parameter solution process, as follows: In the process of constructing the first training data set, for any time t, the collected image data is recorded as F t =[f t (1) ,f t (2) ,…,f t (k) ], where f t (i) , i=1,2,…,k represent the image data collected under k different camera positions and shooting angles; let Represent the damaged area of ​​the belt conveyor surface rubber and the material coverage area respectively; at any time t, the first training sample is represented as As of time t, the first training data set is represented as In the first parameter solving process, for the visual analysis algorithm Ψ, the loss function Loss1 is set, and the optimization goal is to minimize Loss1, and the parameters of the visual analysis algorithm Ψ are solved using the first training data set; the model corresponding to the visual analysis algorithm Ψ includes a first input layer, a shared layer, a specific task layer and a first output layer; wherein the first input layer is based on the training sample Input F i , i=1,2,…,t as input; the shared layer is used to extract the shared features of the input data between different tasks; the specific task layer consists of the rubber damage analysis and the material accumulation analysis; the first output layer is used to output the corresponding input F i ,i=1,2,…,t damaged lagging area And the area of ​​material accumulation on the drum surface and Calculated or estimated by the number of pixels covering the damaged lagging and the accumulated material; the loss function Loss1 is used to measure the true value and predicted values The gap between.

6. The multi-source data collaborative monitoring method according to claim 3, characterized in that: The training of the time series analysis algorithm includes the second training data set construction process and the second parameter solution process, as follows: In the process of constructing the second training data set, for any time t, the collected time series data is recorded as S t =[A t ,N t ,E t ,I t ], where A t Indicates vibration data, N t represents noise data, E t Indicates temperature data, I t Indicates current data; As the belt conveyor runs, the data set formed after data collection is represented as {S1, S2, …, S t }, let the linear sliding window size be w, then the input at time t-1 is represented as {S t-1 ,S t-2 ,…,S t-w }; The second training sample corresponding to time t is expressed as As of time t, the second training data set is represented as In the second parameter solving process, for the time series analysis algorithm Φ, the loss function Loss2 is set, and the optimization goal is to minimize Loss2. The parameters of the time series analysis algorithm Φ are solved using the second training data set. The model corresponding to the time series analysis algorithm Φ includes a second input layer, a feature extraction layer, a fusion layer, and a second output layer. The second input layer is based on the training samples. Input {S i-1 ,S i-2 ,…,S i-w }, i = w, w + 1, ..., t as input; the feature extraction layer includes vibration feature extraction, noise feature extraction, temperature feature extraction and current feature extraction, which are used to extract features of different types of data respectively; the fusion layer is used to splice the four features; The second output layer is used to output the output corresponding to the input {S i-1 ,S i-2 ,…,S i-w },i=w,w+1,…,t’s predicted value, which can be expressed as in Each parameter is a predicted value corresponding to the collected time series data; The loss function Loss2 is used to measure the true value S i and predicted values The gap between.

7. The multi-source data collaborative monitoring method according to claim 1, characterized in that: In the collaborative monitoring step of S4, for each time t, the speed analysis, visual analysis, and time series analysis are performed in parallel, and their respective analysis results are output; In the speed analysis, the speed of the driving drum and the driven drum is analyzed to determine whether slippage occurs, and then determine whether the driving drum is in a normal state, the rubber lagging is damaged, or the bearing is damaged; In visual analysis, the image of the lagging surface is analyzed to determine whether there is material accumulation and whether the lagging is damaged, and then determine whether the drive pulley is in a normal state or a damaged state, and at the same time predict whether the lagging will be damaged; In time series analysis, vibration data, noise data, temperature data and current data are analyzed collaboratively to perform single-step judgment and future trend analysis, and output whether a fault has occurred and a prediction of future fault trends.

8. The multi-source data collaborative monitoring method according to claim 7, characterized in that: Speed ​​analysis includes bearing fault monitoring and lagging fault monitoring; In bearing fault monitoring: collect the speed data of the transmission roller shaft and the driven roller shaft, set the rated speed and the transmission roller speed fluctuation range; determine the transmission roller speed based on the collected data and the set data; when the speed is determined to be lower than the preset threshold, output the judgment result of "bearing damage"; when the speed is determined to be within the preset normal range, perform bearing trend analysis; In the bearing trend analysis, by presetting the abnormal trend threshold, it is determined whether the transmission roller has a trend of increasing rotation resistance; if so, it is predicted that "the bearing may fail", the prediction result is output, and the staff is reminded to check the bearing failure; In the monitoring of rubber lagging faults: set the speed difference fluctuation range and calculate it. If the calculation result is greater than the preset threshold, it is determined that the speed of the driven roller is lower than the speed of the driving roller, indicating that there is a slip problem, and the "rubber lagging may be damaged" is predicted. The prediction result is output and the staff is reminded to troubleshoot the rubber lagging fault; if the calculation result is within the preset normal range, the rubber lagging trend analysis is performed; In the lagging trend analysis, by presetting the abnormal trend threshold, it is determined whether there is a slipping trend; if so, the prediction result of "slipping may occur" is output and the staff is reminded to check the failure of the lagging damage on the surface of the transmission roller; otherwise, the monitoring result of "no obvious abnormality" is output.

9. The multi-source data collaborative monitoring method according to claim 7, characterized in that: Visual analysis is based on the collected data and uses visual analysis algorithms to perform rubber lagging damage analysis and material accumulation analysis, outputting the rubber lagging damage area and the material accumulation area on the drive drum surface; The specific process of rubber bag damage analysis is as follows: set the rubber bag damage judgment threshold, calculate it in combination with the surface area of ​​the transmission roller, and if the calculation result does not meet the rubber bag damage judgment threshold range, output the judgment result of "rubber bag damage"; Otherwise perform lagging trend analysis; In the rubber lagging trend analysis, by presetting the abnormal trend threshold, it is judged whether there is a trend of rubber lagging damage or increasing shedding area; if so, it is predicted that "the rubber lagging surface may be damaged", the prediction result is output and the staff is reminded to check the fault of rubber lagging damage on the surface of the transmission roller; The specific process of material accumulation analysis is as follows: set the material accumulation judgment threshold, calculate and determine whether the area of ​​material accumulation on the surface of the transmission roller is too much, and if so, output the prediction result and remind the staff to handle it; Otherwise, perform material trend analysis, determine whether there is a trend of material accumulation by presetting abnormal trend thresholds, and output the corresponding prediction results.

10. The multi-source data collaborative monitoring method according to claim 7, characterized in that: The specific process of time series analysis is as follows: based on the collected data, the time series analysis algorithm is used to extract the features of vibration data, noise data, temperature data and current data respectively. After calculation and processing, the predicted values ​​corresponding to vibration, noise, temperature and current are output; the abnormal judgment thresholds of vibration, noise, temperature and current are set respectively, and the calculated predicted values ​​are compared with the four types of abnormal judgment thresholds; if any of the predicted values ​​does not meet the abnormal judgment threshold of the corresponding category, it is determined that the transmission roller bearing is running abnormally, and the judgment result of "bearing fault" is output; Otherwise, abnormal trend analysis is performed; In the abnormal trend analysis, the abnormal trend thresholds of vibration, noise, temperature and current are set respectively. After the abnormal trend corresponding to each category is calculated and a judgment vector is formed, it is compared with the reference vector formed corresponding to the set abnormal trend threshold; If the judgment vector of any category does not meet any threshold element in its corresponding reference vector, it is judged that there is a trend of increasing values ​​in the corresponding category, and it is predicted that "the bearing operation may be abnormal", the prediction result is output and the staff is reminded to check the bearing fault; otherwise, the monitoring result of "no obvious abnormality" is output.

Images