Coal mine mechanical fault diagnosis method fusing AI and multivariate information analysis, electronic equipment and medium

Abstract

The invention discloses a coal mine machinery fault diagnosis method fusing AI and multivariate information analysis, electronic equipment and a medium. The method comprises the following steps: acquiring working data of a plurality of different types of coal mine machinery; for each type of coal mine machinery, constructing a data set according to the corresponding working data to train a plurality of machine learning models so as to identify whether the coal mine machinery breaks down or not, and sorting the adaptive priorities of the plurality of machine learning models according to a training result; and for each type of coal mine machinery, analyzing and judging the real-time working data of the coal mine machinery by adopting the at least one machine learning model in the sequence of the adaptive priorities to obtain a fault identification result. According to the method, comprehensive feature extraction can be carried out, the applicability and accuracy between the model and the equipment are higher, and fault judgment of various coal mine machines can be rapidly and accurately carried out.

Description

Technical Field

[0001] The present application relates to the field of fault diagnosis and analysis, and in particular to a coal mine machinery fault diagnosis method, electronic equipment and medium that integrates AI and multivariate information analysis. Background Art

[0002] With the development of science and technology and the advancement of manufacturing technology, the complexity of mechanical equipment in industrial applications continues to increase. Any abnormality or failure during use not only directly affects the use of the product, but may also cause serious safety accidents. As the core equipment of coal mine production, the stable and reliable operation of underground electromechanical equipment in coal mines is the premise for ensuring the safety and economic benefits of coal mining enterprises. Failure to accurately determine the failure and type of machinery in a timely manner will lead to serious accidents. The vibration signal of coal mine machinery includes its own state characteristics and fault information. Analyzing the vibration signal of mechanical equipment is the focus of fault diagnosis. Fault diagnosis methods can detect potential mechanical failures and identify specific types of failures. This active diagnosis method helps reduce maintenance costs and prevent accidents.

[0003] The core of intelligent fault diagnosis is to generate an entity that can perform fault diagnosis on the equipment as an "expert" and give the same diagnostic results as traditional expert detection. At present, model-based fault diagnosis methods are the first to be developed. With the continuous advancement of technology, the objects to be tested are becoming more complex, large-scale, nonlinear, and systematic. It is becoming increasingly difficult to establish an accurate mathematical model. The coupling between various devices makes the model difficulty increase exponentially.

[0004] Chinese patent CN202311168065.2 introduces a "method and device for constructing a device fault diagnosis model based on multi-source feature extraction". The method mainly includes the following steps: first, obtain a pre-constructed sample set; second, use these sample labels to represent different fault types of the equipment; then, extract features from the sample set to obtain target features based on time series features; finally, input these target features and their corresponding sample labels into a preset classifier, train the classifier through a classification learning process, and determine the final fault diagnosis model when specific conditions are met. However, it is worth noting that this method does not fully consider the complexity of the vibration signal itself, and relying solely on time series features may not be sufficient to cope with various challenges in practical applications. In addition, there are certain limitations to using a single model for fault classification, which may limit the flexibility and accuracy of the model in identifying different fault modes. Therefore, in order to further improve the effect of fault diagnosis, it is possible to explore combining more types of features and using methods such as ensemble learning to enhance the robustness and generalization ability of the model.

[0005] Chinese patent CN118964881A introduces a temperature and vibration data analysis system for mine equipment, which consists of several key components: data acquisition module, data preprocessing module, data analysis module, data alarm module, data storage module and analysis data output module. By installing temperature and vibration sensors on the key components of each device, the system can monitor the temperature and vibration intensity of these locations in real time, and then evaluate the overall operating status of the equipment. Once an abnormal situation is detected, the system can immediately issue an alarm and notify the user to take corresponding measures in a timely manner. Although this technology provides important support for improving mine safety and equipment reliability, it mainly relies on the experience of experts to conduct fault analysis.

[0006] In summary, the current response methods used in the field of equipment condition monitoring and fault diagnosis have the following limitations:

[0007] (1) Limited feature extraction dimensions: Most existing technologies tend to focus on a single type of feature extraction when evaluating equipment status, such as limited to time domain features or frequency domain features. Although this approach can reflect the operating status of the equipment to a certain extent, it lacks comprehensive consideration of multi-dimensional features (such as time-frequency domain, phase space reconstruction features, etc.), making it difficult to provide a comprehensive and in-depth status evaluation.

[0008] (2) Insufficient flexibility of the classification model: The fault classification model is single and has low applicability, making it difficult to efficiently select and extract features from equipment fault data. This makes it difficult for the algorithm to accurately identify and classify the data, which limits its ability to adapt to different application scenarios.

[0009] (3) Dealing with imbalanced and small sample data is challenging: Many traditional methods show obvious limitations when encountering severe skewed category distribution or extremely limited sample size. Such problems not only increase the risk of false positives and missed detections, but also make the model training process more difficult. Summary of the invention

[0010] The embodiments of the present application provide a coal mine machinery fault diagnosis method, electronic device and medium that integrate AI and multivariate information analysis to solve the above-mentioned problems existing in the prior art.

[0011] According to one aspect of the present application, a coal mine machinery fault diagnosis method integrating AI and multivariate information analysis is provided, comprising:

[0012] Collect working data of multiple different types of coal mining machinery;

[0013] For each type of coal mining machinery, a data set is constructed based on its corresponding working data to train multiple machine learning models to identify whether the coal mining machinery has failed, and the adaptation priorities of the multiple machine learning models are sorted according to the training results;

[0014] For each type of coal mining machinery, at least one machine learning model in the order of the adaptation priority is used to analyze and judge the real-time working data of the coal mining machinery to obtain a fault identification result.

[0015] Furthermore, the method of constructing a data set based on the corresponding working data to train multiple machine learning models to identify whether the coal mining machinery fails includes:

[0016] A generative adversarial network is used to expand the working data.

[0017] Furthermore, the method of constructing a data set based on the corresponding working data to train multiple machine learning models to identify whether the coal mining machinery fails includes:

[0018] The data set includes time series features, frequency domain features and time-frequency domain features.

[0019] Furthermore, the time-frequency domain features are obtained by one or more of short-time Fourier transform, wavelet transform and Hilbert transform.

[0020] Furthermore, different types of coal mining machinery correspond to different time-frequency domain feature combinations. For a certain type of coal mining machinery, its corresponding time-frequency domain feature combination is obtained by the following method:

[0021] Determine a number of preset time-frequency domain features corresponding to the current type of coal mining machinery, and define the remaining time-frequency domain features other than the number of preset time-frequency domain features as time-frequency domain features to be selected;

[0022] Traversing at least one of the to-be-selected time-frequency domain features and all the preset time-frequency domain features to form a first time-frequency domain feature combination, and using the first time-frequency domain feature combination to participate in training;

[0023] Obtaining, according to the training results, a validity sequence of the time-frequency domain features under each machine learning model, and selecting a plurality of top-ranked ones from the validity sequence as preferred time-frequency domain features;

[0024] The above-mentioned preferred time-frequency domain feature combinations are traversed to form a second time-frequency domain feature combination, and the second time-frequency domain feature combination is used to participate in the analysis and judgment of the real-time working data of the coal mining machinery, and the adaptation relationship between the machine learning model and the second time-frequency domain feature combination is obtained according to the judgment result, and the second time-frequency domain feature combination with the best adaptation relationship is matched with the machine learning model for analysis and judgment.

[0025] Furthermore, the method of constructing a data set based on the corresponding working data to train multiple machine learning models to identify whether the coal mining machinery fails includes:

[0026] The plurality of machine learning models include support vector machines, random forests, K nearest neighbor networks, naive Bayesian networks, extreme gradient boosting, and long short-term memory networks;

[0027] The grid search method is used to determine the parameters of each machine learning model.

[0028] Further, when the coal mining machinery is a bearing or a gearbox, the machine learning model is preferably selected as a naive Bayesian network or a random forest;

[0029] When the coal mining machinery is a motor or a coupling, the machine learning model is preferably selected as a support vector machine or extreme gradient boosting.

[0030] Furthermore, the long short-term memory network uses a softmax activation function as an output layer and a cross entropy loss as a loss function.

[0031] A second aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the electronic device implements the method described in the first aspect of the present invention.

[0032] According to a third aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium is used to store a computer program. When the computer program is executed on a computer, the computer executes the method according to the first aspect of the present invention.

[0033] Beneficial Effects

[0034] The present invention provides an advanced coal mine machinery fault diagnosis method, which deeply integrates artificial intelligence and multivariate information analysis technology, thereby achieving efficient management of the health status of the equipment. Not only its time domain characteristics and frequency characteristics are extracted, but also the hidden information in the time and frequency domains is deeply mined to ensure that the most subtle abnormal changes are captured. Based on these rich feature information, fault models for various key components such as motors, bearings, gearboxes, etc. are constructed, and classic machine learning algorithms such as support vector machines and deep learning architectures such as long short-term memory networks are used to predict and classify faults. It can accurately output the fault category and cause, and can also provide targeted solution suggestions to help the maintenance team respond quickly and effectively deal with problems.

[0035] Compared with the prior art, the solution proposed by the present invention has significant technical advantages in many aspects, which are specifically reflected in the following key points:

[0036] (1) Comprehensive feature extraction: Unlike traditional methods that rely only on a single type of data feature (such as time domain or frequency domain features), the present invention uses time domain features, frequency domain features, and time-frequency domain features to fully mine the potential information in the data. This fusion of multi-dimensional features helps to more deeply understand the essential properties of the signal and its changing laws, thereby providing richer and more accurate basic information for subsequent fault identification.

[0037] (2) Enhance model applicability and accuracy: For the analysis of equipment failure types, the present invention adopts a combination of multiple machine learning algorithms. Compared with relying solely on a certain model, this method can not only effectively improve the learning ability for complex failure modes, but also overcome the overfitting problem to a certain extent, and enhance the generalization performance of the model for different types of failures. In addition, by combining the advantages of multiple learning mechanisms, the system can achieve higher diagnostic accuracy and ensure the reliability of fault detection results.

[0038] (3) Solve the problem of sample imbalance: This paper introduces generative adversarial network technology to expand the sample size of rare fault categories. In this way, it can not only effectively alleviate the model bias caused by uneven sample distribution, but also significantly improve the recognition rate of those rare but important fault types. More importantly, this method avoids the situation where some critical faults are misjudged as normal states, greatly reducing the occurrence of underreporting. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0040] Figure 1 It is a flow chart of a coal mine machinery fault diagnosis method integrating AI and multivariate new analysis according to an embodiment of the present application.

[0041] Figure 2 It is a schematic diagram of real-time warning of single-point data according to an embodiment of the present application.

[0042] Figure 3 It is a data processing flow chart according to an embodiment of the present application.

[0043] Figure 4 It is a feature extraction flow chart according to an embodiment of the present application.

[0044] Figure 5It is a schematic diagram of the feature formation process of different coal mining machinery according to the embodiment of the present application.

[0045] Figure 6 It is a schematic diagram of the characteristic combination of the motor under different models according to the embodiment of the present application.

[0046] Figure 7 It is a schematic diagram of the characteristic combination of the bearing under different models according to the embodiment of the present application.

[0047] Figure 8 It is a schematic diagram of the characteristic combination of the gearbox under different models according to the embodiment of the present application.

[0048] Fig. 9 It is a schematic diagram of the feature combination of the coupling under different models according to the embodiment of the present application.

[0049] Fig.10 It is a schematic diagram of determining model parameters according to the grid search method of an embodiment of the present application.

[0050] Fig.11 It is a flowchart of fault diagnosis and alarm push according to an embodiment of the present application.

[0051] Fig.12 It is a schematic diagram of a coal mine machinery fault diagnosis device that integrates AI and new multivariate analysis according to an embodiment of the present application.

[0052] Fig.13 is a schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0053] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0054] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0055] An embodiment of the present invention provides an efficient coal mine machinery fault diagnosis method that integrates AI and multivariate information analysis. In order to achieve more comprehensive and accurate detection of potential faults of underground equipment, in this embodiment, the coal mine machinery includes four types: motors, bearings, couplings and gear boxes.

[0056] The motor is a power device that converts electrical energy into mechanical energy. It is mainly used to drive various machines and systems in coal mining equipment. For example, scraper conveyors, belt conveyors, hoists, etc. all rely on motors to provide power. There are five main types of faults: motor rotor imbalance, angle misalignment, comprehensive misalignment, loose motor feet, and loose motor structure.

[0057] The main function of the bearing is to support the mechanical rotating body, reduce the friction coefficient during its movement, and ensure its rotation accuracy. There are seven main types of faults: peeling of the inner ring of the bearing, peeling of the outer ring of the bearing, peeling of the rolling element of the bearing, loose rotation of the inner and outer rings of the bearing, scratching of the cage, fracture of the cage, and bearing collision and wear.

[0058] The coupling is used to connect two shafts so that they rotate together and transmit torque. It can compensate for the misalignment error between the two shafts to a certain extent and play a protective role. There are mainly three types of faults: coupling parallel misalignment, coupling wear, and coupling tear.

[0059] Gearboxes are widely used in the reduction transmission system of coal mining machinery, such as belt conveyors, crushers, etc. There are mainly six types of faults: unbalanced gears on the high-speed shaft of the gearbox, poor gear meshing, eccentric gear wear, broken gears, loose gearbox feet, and loose gearbox structure.

[0060] In this embodiment, for each type of equipment, not only is a detailed classification of fault types performed, but also the specific causes that may cause each type of fault are deeply analyzed, thereby laying a foundation for subsequent rapid and accurate location of the root cause of the problem.

[0061] like Figure 1 As shown in FIG. 1 , a flow chart of an efficient coal mine machinery fault diagnosis method integrating AI and multivariate information analysis is shown in FIG. Figure 1 It can be seen that the method comprises the following steps:

[0062] Step S101: Collect working data of multiple different types of coal mining machinery.

[0063] Step S102: For each type of coal mining machinery, a data set is constructed based on its corresponding working data to train multiple machine learning models to identify whether the coal mining machinery has a fault, and the adaptation priorities of the multiple machine learning models are sorted according to the training results.

[0064] Step S103: For each type of coal mining machinery, at least one machine learning model in the order of the adaptation priority is used to analyze and judge the real-time working data of the coal mining machinery to obtain a fault identification result.

[0065] The above method can improve the feature quality through multivariate information analysis and construct multiple classification models to ensure the accuracy of fault diagnosis.

[0066] In this embodiment, a comprehensive model diagnosis library is constructed for each type of equipment. The library includes multiple algorithms to cover different types of fault characteristics and improve the overall performance of fault diagnosis. Specifically, this model library includes the following methods:

[0067] (1) Traditional machine learning model:

[0068] Support Vector Machine (SVM): Applicable to high-dimensional data, it distinguishes samples of different categories by finding the optimal hyperplane.

[0069] Random Forest (RF): An ensemble learning method based on a set of decision trees that can handle a large number of input variables and has good resistance to overfitting.

[0070] K Nearest Neighbor (KNN): Predict the category of an unknown sample based on the majority category of its nearest neighbors.

[0071] Naive Bayesian Network: A probabilistic classifier based on Bayes' theorem.

[0072] (2) Ensemble learning model:

[0073] Extreme Gradient Boosting (XGBoost): An efficient implementation of the gradient boosting framework that iteratively adds new weak learners to minimize the loss function.

[0074] (3) Deep Learning Model:

[0075] Long Short-Term Memory (LSTM): is a special type of recurrent neural network (RNN) that is able to learn long-term dependencies and was originally designed to process and predict important features in time series data. Due to its powerful properties, LSTM can be used not only for regression tasks but also for classification tasks.

[0076] To apply it to classification problems, the following aspects need to be adjusted:

[0077] Output layer adjustment: For classification tasks, modify the output layer of the LSTM network to adapt to the classification requirements. Use the softmax activation function as the output layer, and set the number of neurons in this layer to be equal to the number of categories. Since the number of faults of motors, bearings, etc. is inconsistent, it is necessary to make modifications based on the actual situation;

[0078] Loss function selection: Select the cross entropy loss function.

[0079] Training goal: During the training process, the loss function is minimized through the back-propagation algorithm, thereby adjusting the weights of the LSTM network so that it can better classify the input.

[0080] LSTM classification has the following advantages:

[0081] Ability to process long sequence data: In application scenarios such as vibration signal analysis, data is often continuous and large time series. LSTM network has the inherent advantage of processing such long-span, multi-sequence data. It can effectively capture and learn long-term dependencies in the sequence, and maintain stable performance even in the face of a steady stream of new data.

[0082] Architectural flexibility: The LSTM model exhibits a high degree of architectural flexibility, allowing researchers to build more complex network structures by stacking multiple LSTM layers or combining them with other types of layers (such as fully connected layers and convolutional layers). LSTM models can be quickly built and adjusted for different motor, bearing and other data sets, thereby improving development efficiency.

[0083] End-to-end learning mechanism: LSTM can automatically extract features directly from raw data without manual intervention. This approach avoids the risk of missing information that may occur in early feature extraction, ensuring that all potentially important information is taken into account, thereby improving the accuracy of classification results.

[0084] Specifically, the implementation of the above method mainly includes six parts: data collection and real-time warning, data processing, feature analysis, model training, fault diagnosis and alarm push.

[0085] Part 1: Data Collection and Real-time Warning

[0086] like Figure 2 As shown, a vibration information collection module is set on the coal mine machinery of interest to collect the vibration data of the equipment and display it in the form of vibration acceleration. In order to ensure the measurement accuracy, this embodiment adopts a high-sensitivity three-axis sensor to record the vibration conditions in three orthogonal directions of X, Y, and Z respectively. This three-dimensional vibration data collection method enables the system to understand the dynamic behavior of the equipment more comprehensively. Based on the collected acceleration sequence, the system calculates the vibration intensity using the gradient integration method. Finally, the calculated vibration intensity is strictly evaluated according to ISO international standards. This process involves comparing the actual measurement value with the limit value specified in the standard to determine whether the equipment is within the normal working range and whether there is an abnormal situation, and four evaluation criteria of "normal", "warning", "alarm" and "fault" are given.

[0087] Part II: Data Processing

[0088] like Figure 3 As shown, a predefined sample set is constructed based on the collected data. The sample set includes vibration signals and their corresponding sample labels, wherein the sample labels are used to represent the specific fault type of the equipment.

[0089] First, a local outlier factor (LOF) algorithm is used to identify and remove anomalous data points. This preprocessing step is critical to ensure data quality during subsequent model training and helps improve the accuracy and reliability of the final model.

[0090] Next, in order to solve the sample imbalance problem in the data set, the generative adversarial network (GAN) technology was introduced to expand the data set. In particular, when there are insufficient samples of the fault type, additional synthetic samples are generated through GAN so that the number of fault samples can be close to or reach a level equivalent to the number of normal samples. This method not only increases the diversity of the data set, but also ensures that the model can learn the feature differences between different categories more evenly during the training process, thereby significantly improving the model's recognition ability for minority categories and overall performance.

[0091] Through the above method, the robustness of the fault diagnosis system can be effectively enhanced, so that it has stronger data adaptability and higher diagnostic accuracy when facing complex and changeable actual application scenarios.

[0092] Part III: Feature Analysis

[0093] like Figure 4 As shown, in an embodiment, the data set is constructed based on the corresponding working data to train multiple machine learning models to identify whether the coal mining machinery fails, including:

[0094] The data set includes time series features, frequency domain features and time-frequency domain features.

[0095] For different coal mining machines, the feature combinations of the corresponding data sets are different. Regardless of the type of coal mining machine, this embodiment divides the feature parameters of the data set into two parts: time series features and non-time series features. Figure 5 As shown in the figure, for the time series feature part, multiple time domain feature indicators including mean, standard deviation, root amplitude, peak-to-peak value, skewness, kurtosis, peak factor, margin factor, etc. are selected to fully capture the time dynamic characteristics of the data. In addition, by applying a variety of technical means such as Fourier transform, power spectral density analysis and inverse spectrum analysis, the time series characteristics are further analyzed from the frequency perspective to obtain their corresponding frequency domain characteristics. On this basis, the time-frequency domain characteristics are obtained through one or more of short-time Fourier, wavelet transform and Hilbert transform, and the time-frequency domain characteristics are used to accurately describe complex dynamic behaviors, so that the model can capture the dynamic process of subtle changes under different working conditions.

[0096] In modern signal processing technology, time domain and frequency domain analysis methods are widely used due to their relatively simple parameter requirements. These traditional methods rely on a small number of key parameters and can often obtain satisfactory analysis results through empirical rules. However, for the time-frequency domain, short-time Fourier transform and wavelet transform need to determine more parameters.

[0097] like:

[0098] The main parameters of short-time Fourier transform include: window function, window length, overlap rate, and FFT size;

[0099] The main parameters of wavelet transform: mother wavelet, scale, translation, number of decomposition levels, and threshold processing.

[0100] These characteristic parameters have a significant impact on different types of signals. Therefore, how to reasonably determine these parameters according to the specific signal type becomes the key to applying time-frequency domain analysis.

[0101] Therefore, in order to more accurately determine the key parameters in the time-frequency domain analysis, in this embodiment, different types of coal mining machinery correspond to different time-frequency domain feature combinations. For a certain type of coal mining machinery, its corresponding time-frequency domain feature combination is obtained by the following method:

[0102] Step S201: determine a number of preset time-frequency domain features corresponding to the current type of coal mining machinery, and define the remaining time-frequency domain features other than the preset time-frequency domain features as time-frequency domain features to be selected.

[0103] The preset time-frequency domain features are usually the features that must be considered when judging faults. The above preset time-frequency domain features can be initially set under the guidance of expert experience, including consulting a large amount of literature and consulting senior experts in the industry to initially set the parameters. This step ensures the quality of parameter initialization and also reduces the space for blind search.

[0104] Step S202: traverse at least one of the to-be-selected time-frequency domain features and all the preset time-frequency domain features to form a first time-frequency domain feature combination, and use the first time-frequency domain feature combination for training.

[0105] Step S203: Obtain the validity sequence of the time-frequency domain features under each machine learning model according to the training results, and select several top-ranked ones from the validity sequence as preferred time-frequency domain features.

[0106] Step S204, traverse the above-mentioned preferred time-frequency domain feature combination to form a second time-frequency domain feature combination, and use the second time-frequency domain feature combination to participate in the analysis and judgment of the real-time working data of the coal mine machinery, and obtain the adaptation relationship between the machine learning model and the second time-frequency domain feature combination according to the judgment result, and match the second time-frequency domain feature combination with the best adaptation relationship with the machine learning model for analysis and judgment.

[0107] The above method first presets some parameters based on the deep knowledge and practical experience of experts in the field to ensure that the initial settings can reflect the best practices of signal processing. Next, by applying these preset parameters to many different types of fault diagnosis models, the powerful computing power and rich data resources of the big data platform are used to achieve extensive exploration of various possible parameter configurations.

[0108] By learning from massive amounts of historical data, the system can identify which parameter configurations are most effective under what circumstances, and dynamically adjust parameter values ​​accordingly to adapt to new application scenarios or changing data characteristics.

[0109] As a preferred embodiment, based on the above method, the diagnostic accuracy is further used as the core evaluation criterion, and other relevant performance indicators (such as computing efficiency, robustness, etc.) are considered to comprehensively evaluate the effect of each parameter configuration. After multiple rounds of testing and optimization, the best parameter combination is selected for actual deployment.

[0110] As a preferred embodiment, based on the above method, a closed-loop feedback system is established to allow continuous collection of new data when running in a real environment, further fine-tuning and optimizing parameters, making the entire process a continuous improvement process.

[0111] Through the above method, the scientificity and rationality of feature selection are improved.

[0112] Next, in order to optimize the feature set, improve the model processing efficiency and enhance the model generalization ability, principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) techniques are used to perform feature selection and dimensionality reduction on all extracted features, aiming to reduce the feature dimension while retaining the core information of the data, thereby obtaining a more compact and representative time series feature representation.

[0113] like Figures 6 to 9 As shown, the feature combinations of motors, bearings, gearboxes and couplings under different models are respectively shown. It can be seen that the feature combinations of the same coal mining machinery are different when using different models, and different machinery also corresponds to different feature combinations. Therefore, the method of this embodiment can more efficiently screen out real and effective feature combinations, thereby improving the accuracy and efficiency of recognition.

[0114] Finally, the optimized time series feature representation is deeply integrated with the non-time series features to construct a comprehensive target feature set, which provides a solid foundation for subsequent model training and prediction, and improves the expressiveness of the features and the generalization performance of the model.

[0115] Part 4: Model Training

[0116] like Fig.10 As shown, in order to ensure that the selected model can achieve the best performance, this patent uses the grid search method to carefully tune the key parameters of each model. Grid Search is a technology used for hyperparameter optimization. It searches for the optimal hyperparameter combination by exhaustively enumerating the specified parameter values ​​and automatically selects the best performing settings, thereby avoiding the subjectivity and uncertainty caused by manual parameter adjustment, thereby improving model performance.

[0117] Take random forest as an example to introduce the basic steps of grid search method:

[0118] Define parameter space: First determine the hyperparameters to be optimized and their possible value ranges. Here are some of the main parameters:

[0119] Create a grid: Based on the possible values ​​of the above parameters, create all possible parameter combinations. This forms a "grid" where each point represents a specific set of hyperparameter values.

[0120] Cross-validation evaluation: For each set of parameter combinations in the grid, train the model and evaluate its performance using cross-validation (usually k-fold cross-validation). This process involves splitting the dataset into multiple subsets, using one of the subsets as the test set and the remaining subsets as the training set, and repeating this process until each subset has been used as the test set once. Finally, calculate the average performance score over all folds.

[0121] Select the optimal parameters: Compare the cross-validation scores under all parameter combinations and select the set of parameters with the highest average score as the optimal parameter combination.

[0122] Apply Best Parameters: Retrain the model using the best parameters found and evaluate its performance on an independent test set.

[0123] Similarly, determine the parameters of the models in the model library such as motors and bearings.

[0124] Next, using the extracted time series features and their corresponding sample labels, the data set is divided into two parts: a training set and a test set. 70% of the data is used as a training set to train the various fault diagnosis models mentioned above; the remaining 30% of the data is used as an independent test set to evaluate the generalization ability of each model on unseen data. During the training process, each model learns how to identify different fault modes from given features.

[0125] After training, the test set is used to verify the actual effect of each model, focusing on their classification accuracy on new data. By comparing the performance of different models, the models with the highest accuracy on the test set are selected and will be selected as the final fault classification tool. The purpose of this is to ensure that the fault detection system can not only work well under known conditions, but also effectively deal with new or abnormal conditions that may arise in actual applications.

[0126] Different models can give full play to their respective advantages, ensuring the robustness of the model while also improving the sensitivity of fault detection. In the early stage of deployment, when there are fewer devices and the amount of data is not large, you can choose a machine learning model to quickly diagnose faults;

[0127] In the middle and late stages of deployment, as the number of devices and the amount of data increase, the traditional machine learning model can be selected for rapid classification, and then LSTM can be used for further analysis, and the results of the two can be combined as the final classification result.

[0128] In some preferred embodiments, in order to obtain better fault results, weights are assigned to different models;

[0129] Assume that the classification result of the support vector machine or random forest or K-nearest neighbor or naive Bayesian network is A, the classification result of the LSTM is B, and the final classification result is C, satisfying C=aA+bB. In this embodiment, a higher weight is given to the result of the LSTM, b=0.6, a=0.4.

[0130] In addition, in order to further enhance the robustness and reliability of the system, the k-fold cross-validation technology is introduced to reduce the deviation that may be introduced due to data partitioning. At the same time, the performance of the model in the actual operating environment is continuously monitored, and the training data set and model parameters are regularly updated to keep the system effective in the long term.

[0131] Part V: Troubleshooting

[0132] In the field of industrial equipment monitoring and fault diagnosis, in order to improve the accuracy and reliability of diagnosis, multiple vibration sensors are usually installed at important parts of key equipment. These sensors can capture various vibration signals generated when the equipment is running, and by analyzing these signals, potential problems or failure modes can be effectively identified.

[0133] like Fig.11 As shown in the figure, first, the data collected from each vibration sensor is subjected to multivariate feature extraction to ensure the data quality of the input model; then, the operator selects a trained fault diagnosis model and feeds the preprocessed data into this model as input; finally, after comprehensively considering the information provided by all sensors, the model will give the final diagnosis result.

[0134] Some reference opinions are also given for the selection of on-site models.

[0135] (1) For bearings and gearboxes, which have a high correlation with the signal, Bayesian classifiers and random forests can be used as the priority.

[0136] (2) For motors and couplings, which are most affected by the environment, support vector machines and limit hoists are preferred;

[0137] (3) For complex data that requires more precise analysis, LSTM is preferred.

[0138] Part 6: Alarm Push

[0139] In a complex and high-risk working environment like a coal mine, timely and accurate communication of fault information is crucial to preventing accidents and reducing losses. This embodiment pushes fault alarms through multiple channels to ensure that information can be quickly and effectively transmitted to relevant personnel, so that timely measures can be taken to avoid potential disasters. At the same time, the system gives the cause of the fault and maintenance suggestions, so that a simple cause can be checked before the arrival of professional maintenance personnel. Specifically, this solution adopts the following 4 notification methods:

[0140] (1) WeChat for Business push: Utilize WeChat for Business, a widely used enterprise-level communication tool, to directly send fault details to designated staff or team groups. This method is not only fast and convenient, but also ensures the security and privacy of information.

[0141] (2) Interface pop-up window push: When the system detects an abnormality, a warning window will automatically pop up on the operation interface to intuitively display the fault information and preliminary handling suggestions to the working personnel. This method is particularly suitable for situations that require immediate response.

[0142] (3) Interface voice push: In addition to visual prompts, the system will also issue an alarm through voice broadcast, which can effectively remind even in the case of poor visibility or distracted operator. This is very helpful for improving alertness and speeding up reaction time.

[0143] (4) Alibaba Cloud SMS push: With the help of the stable and efficient SMS service provided by Alibaba Cloud, important information can be sent directly to the mobile phones of relevant persons in charge. This method can break through network restrictions and ensure that key messages can be received no matter where you are.

[0144] The adoption of the above-mentioned diversified notification strategy can not only overcome the technical obstacles that may be encountered by a single communication method (such as insufficient signal coverage), but also enhance the effectiveness of information transmission, ensuring that everyone from management to front-line employees can be informed of equipment status changes in a timely manner and participate in the problem-solving process. In addition, the multi-channel parallel information dissemination mechanism can also help to form a good atmosphere of mutual supervision, reduce the risk of insufficient attention to safety warnings due to personal negligence, and provide a more solid guarantee for mine safety production.

[0145] This embodiment integrates artificial intelligence technology and multi-information analysis methods to achieve more accurate and efficient monitoring of the status of mechanical equipment and fault prediction. It can discover and warn of possible problems at an early stage, greatly improving the reliability and safety of the entire production system.

[0146] Based on various aspects of the method introduced in the above embodiment, the specific implementation on site is as follows:

[0147] 1. When new data arrives, the vibration acceleration is converted into velocity, and whether there is any abnormality is quickly determined based on ISO standards;

[0148] 2. In order to clearly see the effect of the vibration sequence under multiple spectra on site, pictures are displayed in the time domain, frequency domain and time-frequency domain;

[0149] 3. If the on-site operator wants to check whether the equipment has a fault, he can select a model from the diagnostic model library for diagnosis, or select multiple models for diagnosis and compare the diagnosis results;

[0150] 4. If a failure occurs, push it through multiple methods.

[0151] Based on the same inventive concept as the above method embodiment, the present application embodiment also provides a coal mine machinery fault diagnosis device integrating AI and multivariate information analysis, such as Fig.12 The figure shows a schematic diagram of a coal mine machinery fault diagnosis device integrating AI and multivariate information analysis. As can be seen from the figure, the device includes:

[0152] The acquisition module is used to collect the working data of multiple different types of coal mining machinery;

[0153] A training module, for each type of coal mining machinery, constructing a data set according to its corresponding working data to train multiple machine learning models to identify whether the coal mining machinery has a fault, and sorting the adaptation priorities of the multiple machine learning models according to the training results;

[0154] The judgment module is used to analyze and judge the real-time working data of each type of coal mining machinery using at least one machine learning model in the order of the adaptation priority to obtain a fault identification result.

[0155] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0156] Based on the same inventive concept as the above-mentioned method embodiment, an electronic device is also provided in an embodiment of the present application, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the electronic device implements the control method in the above-mentioned embodiment.

[0157] In one embodiment, the electronic device may be a server. In this embodiment, the structure of the electronic device may be as follows: Figure 3 As shown, it includes a memory 2001 , a communication module 2003 and one or more processors 2002 .

[0158] The memory 2001 is used to store computer programs executed by the processor 2002. The memory 2001 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and programs required for running the instant messaging function, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0159] The memory 2001 may be a volatile memory, such as a random-access memory (RAM); the memory 2001 may also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or the memory 2001 may be any other medium that can be used to carry or store a desired computer program in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2001 may be a combination of the above memories.

[0160] The processor 2002 may include one or more central processing units (CPU) or a digital processing unit, etc. The processor 2002 is used to implement the above-mentioned audio data processing method when calling the computer program stored in the memory 2001 .

[0161] The communication module 2003 is used to communicate with terminal devices and other servers.

[0162] The specific connection medium between the memory 2001, the communication module 2003 and the processor 2002 is not limited in the embodiment of the present application. Fig.13 In the embodiment, the memory 2001 and the processor 2002 are connected via a bus 2004. The bus 2004 is Figure 3 The arrows in the figure are used to describe the connection between other components, which is only for illustrative purposes and is not intended to be limiting. Bus 2004 can be divided into address bus, data bus, control bus, etc. For ease of description, Fig.13 The diagram shows that only one arrow is used, but this does not mean that there is only one bus or only one type of bus.

[0163] Based on the same inventive concept as the above method embodiment, an embodiment of the present invention also provides a computer-readable storage medium, the computer-readable storage medium is used to store a computer program, when the computer program is run on a computer, the electronic device implements the control method in the above embodiment. The computer-readable storage medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0164] Based on the same inventive concept as the above-mentioned method embodiment, an embodiment of the present invention further provides a computer program product, the computer program product including a computer program. When the program product is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of the control method according to various exemplary embodiments of the present application described above in this specification. The program product may adopt any combination of one or more readable media. These computer program commands may be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the commands executed by the processor of the computer or other programmable data processing device generate commands for implementing the steps in the process Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0165] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

Claims

1. A coal mine machinery fault diagnosis method integrating AI and multivariate information analysis, characterized in that: include: Collect working data of multiple different types of coal mining machinery; For each type of coal mining machinery, a data set is constructed based on its corresponding working data to train multiple machine learning models to identify whether the coal mining machinery has failed, and the adaptation priorities of the multiple machine learning models are sorted according to the training results; For each type of coal mining machinery, at least one machine learning model in the order of the adaptation priority is used to analyze and judge the real-time working data of the coal mining machinery to obtain a fault identification result.

2. The method according to claim 1, characterized in that: The step of constructing a data set based on the corresponding working data to train multiple machine learning models to identify whether the coal mining machinery fails includes: A generative adversarial network is used to expand the working data.

3. The method according to claim 1, characterized in that: The step of constructing a data set based on the corresponding working data to train multiple machine learning models to identify whether the coal mining machinery fails includes: The data set includes time series features, frequency domain features and time-frequency domain features.

4. The method according to claim 3, characterized in that: The time-frequency domain features are obtained by one or more of short-time Fourier transform, wavelet transform and Hilbert transform.

5. The method according to claim 4, characterized in that: Different types of coal mining machinery correspond to different time-frequency domain feature combinations. For a certain type of coal mining machinery, its corresponding time-frequency domain feature combination is obtained in the following way: Determine a number of preset time-frequency domain features corresponding to the current type of coal mining machinery, and define the remaining time-frequency domain features other than the number of preset time-frequency domain features as time-frequency domain features to be selected; Traversing at least one of the to-be-selected time-frequency domain features and all the preset time-frequency domain features to form a first time-frequency domain feature combination, and using the first time-frequency domain feature combination to participate in training; Obtaining, according to the training results, a validity sequence of the time-frequency domain features under each machine learning model, and selecting a plurality of top-ranked ones from the validity sequence as preferred time-frequency domain features; The above-mentioned preferred time-frequency domain feature combinations are traversed to form a second time-frequency domain feature combination, and the second time-frequency domain feature combination is used to participate in the analysis and judgment of the real-time working data of the coal mining machinery, and the adaptation relationship between the machine learning model and the second time-frequency domain feature combination is obtained according to the judgment result, and the second time-frequency domain feature combination with the best adaptation relationship is matched with the machine learning model for analysis and judgment.

6. The method according to claim 1, characterized in that: The step of constructing a data set based on the corresponding working data to train multiple machine learning models to identify whether the coal mining machinery fails includes: The plurality of machine learning models include support vector machines, random forests, K nearest neighbor networks, naive Bayesian networks, extreme gradient boosting, and long short-term memory networks; The grid search method is used to determine the parameters of each machine learning model.

7. The method according to claim 6, characterized in that When the coal mining machinery is a bearing or a gearbox, the machine learning model is preferably a naive Bayesian network or a random forest; When the coal mining machinery is a motor or a coupling, the machine learning model is preferably selected as a support vector machine or extreme gradient boosting.

8. The method according to claim 6, characterized in that The LSTM network uses a softmax activation function as the output layer and a cross entropy loss as the loss function.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the electronic device implements the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and when the computer program is run on a computer, the computer is caused to execute the method according to any one of claims 1 to 8.

Images