Equipment fault diagnosis method and device, model training method, equipment and medium

Abstract

The invention relates to the technical field of computers, and discloses an equipment fault diagnosis method and device, a model training method, equipment and a medium. Firstly, target waveform data of to-be-detected equipment is acquired; next, feature extraction is carried out on the target waveform data through the equipment fault diagnosis model, and time domain features and frequency domain features are obtained; wherein the equipment fault diagnosis model comprises a semi-supervised learning module. The parameters of the equipment fault diagnosis model are updated through the semi-supervised learning module by using the unlabeled data and the labeled data, the dynamic change of the equipment state is adapted, the dependence on data labeling is effectively reduced, the training cost is reduced, and the generalization performance of the model is improved; and the equipment fault diagnosis model performs equipment fault diagnosis based on the time domain feature, the frequency domain feature, the time domain threshold range and the frequency domain threshold range to obtain an equipment diagnosis result. By realizing automatic and intelligent fault diagnosis, the accuracy and efficiency of a diagnosis result are improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, model training method, equipment, and medium for diagnosing equipment faults. Background Technology

[0002] The equipment is widely used in various fields, such as manufacturing, healthcare, transportation, and energy, and its applications cover a variety of functions, including production process monitoring, data acquisition, environmental monitoring, and automated control. Therefore, fault diagnosis of the equipment is particularly important.

[0003] In related technologies, equipment fault diagnosis relies on human experience. However, the manual approach is affected by subjective factors, leading to room for improvement in the accuracy and reliability of the diagnostic results. Summary of the Invention

[0004] The embodiments described in this specification aim to at least partially solve one of the technical problems in the related art. To this end, the embodiments described in this specification propose a method, apparatus, model training method, equipment, and medium for diagnosing equipment faults.

[0005] This specification provides a method for diagnosing equipment faults, the method comprising:

[0006] Acquire the target waveform data of the device under test;

[0007] The target waveform data is subjected to feature extraction by the equipment fault diagnosis model to obtain time domain features and frequency domain features. The equipment fault diagnosis model includes a semi-supervised learning module, which is used to update the parameters of the equipment fault diagnosis model.

[0008] The equipment fault diagnosis model performs equipment fault diagnosis based on the time-domain features, the frequency-domain features, the time-domain threshold range, and the frequency-domain threshold range, and obtains the equipment diagnosis results.

[0009] In one implementation, the step of extracting features from the target waveform data using a device fault diagnosis model to obtain time-domain and frequency-domain features includes:

[0010] The target waveform data is divided into waveform segment data corresponding to multiple time windows;

[0011] The vibration frequency features are extracted from the waveform segment data corresponding to the multiple time windows using the equipment fault diagnosis model to obtain the frequency domain features.

[0012] The vibration amplitude features are extracted from the waveform segment data corresponding to the multiple time windows using the equipment fault diagnosis model to obtain the time domain features.

[0013] In one implementation, the equipment fault diagnosis model is obtained in the following manner:

[0014] Construct a set of device waveform data samples;

[0015] Based on the semi-supervised learning module, the parameters of the device fault diagnosis model to be trained are updated using the device waveform data sample set, thereby obtaining the device fault diagnosis model.

[0016] In one implementation, the device waveform data sample set includes an unlabeled waveform data sample set and a labeled waveform data sample set; the semi-supervised learning module uses the device waveform data sample set to update the parameters of the device fault diagnosis model to be trained, thereby obtaining the device fault diagnosis model, includes:

[0017] The fault diagnosis model of the device to be trained is pre-trained using the unlabeled waveform data sample set, and the parameters of the fault diagnosis model of the device to be trained are updated to obtain an intermediate device fault diagnosis model.

[0018] The intermediate equipment fault diagnosis model is trained using the labeled waveform data sample set, and the parameters of the intermediate equipment fault diagnosis model are updated to obtain the equipment fault diagnosis model.

[0019] In one implementation, the construction of the device waveform data sample set includes:

[0020] Use sensors to acquire an initial set of device waveform data;

[0021] The initial device waveform data set is cleaned and normalized to obtain the target waveform data set;

[0022] Data annotation is performed on a portion of the target waveform data in the target waveform data set to obtain the device waveform data sample set.

[0023] In one implementation, the equipment fault diagnosis model is built based on a convolutional neural network (CNN) model.

[0024] This specification provides a model training method for training a device fault diagnosis model according to any one of the above-described embodiments. The model training method includes:

[0025] Construct a set of device waveform data samples;

[0026] Based on the semi-supervised learning module, the parameters of the device fault diagnosis model to be trained are updated using the device waveform data sample set, thereby obtaining the device fault diagnosis model.

[0027] In one implementation, the device waveform data sample set includes an unlabeled waveform data sample set and a labeled waveform data sample set; the semi-supervised learning module uses the device waveform data sample set to update the parameters of the device fault diagnosis model to be trained, thereby obtaining the device fault diagnosis model, includes:

[0028] The fault diagnosis model of the device to be trained is pre-trained using the unlabeled waveform data sample set, and the parameters of the fault diagnosis model of the device to be trained are updated to obtain an intermediate device fault diagnosis model.

[0029] The intermediate equipment fault diagnosis model is trained using the labeled waveform data sample set, and the parameters of the intermediate equipment fault diagnosis model are updated to obtain the equipment fault diagnosis model.

[0030] In one implementation, the construction of the device waveform data sample set includes:

[0031] Use sensors to acquire an initial set of device waveform data;

[0032] The initial device waveform data set is cleaned and normalized to obtain the target waveform data set;

[0033] Data annotation is performed on a portion of the target waveform data in the target waveform data set to obtain the device waveform data sample set.

[0034] In one implementation, the equipment fault diagnosis model is built based on a convolutional neural network (CNN) model.

[0035] This specification provides a device for diagnosing equipment faults, the device comprising:

[0036] The waveform data acquisition module is used to acquire the target waveform data of the device under test;

[0037] The time-frequency domain feature extraction module is used to extract features from the target waveform data through the equipment fault diagnosis model to obtain time-domain features and frequency-domain features. The equipment fault diagnosis model includes a semi-supervised learning module, which is used to update the parameters of the equipment fault diagnosis model.

[0038] The equipment fault diagnosis module is used to perform equipment fault diagnosis based on the time domain features, the frequency domain features, the time domain threshold range, and the frequency domain threshold range, and obtain equipment diagnosis results.

[0039] This specification provides a computer device comprising: a memory, and one or more processors communicatively connected to the memory; the memory stores instructions executable by the one or more processors, the instructions being executed by the one or more processors to cause the one or more processors to perform the steps of the method described in any of the above embodiments.

[0040] This specification provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the above embodiments.

[0041] This specification provides a computer program product that includes instructions that, when executed by a processor of a computer device, enable the computer device to perform the steps of the method described in any of the above embodiments.

[0042] In the above-described implementation method, firstly, target waveform data of the device to be tested is acquired. Next, feature extraction is performed on the target waveform data using a device fault diagnosis model to obtain time-domain and frequency-domain features. The device fault diagnosis model includes a semi-supervised learning module. This module utilizes both unlabeled and labeled data to update the parameters of the device fault diagnosis model, adapting to dynamic changes in the device's state, effectively reducing reliance on data labeling, lowering training costs, and improving the model's generalization ability. Finally, the device fault diagnosis model performs device fault diagnosis based on the time-domain features, frequency-domain features, time-domain threshold range, and frequency-domain threshold range to obtain the device diagnosis result.

[0043] The above-described implementation reduces reliance on human experience through equipment fault diagnosis models, making the identification of equipment fault characteristics faster and more accurate. Furthermore, by achieving automated and intelligent fault diagnosis, the implementation improves the accuracy and efficiency of diagnostic results, reducing maintenance and repair costs. For vibration data under complex fault conditions, it can quickly locate hidden faults, improving the comprehensiveness and timeliness of fault diagnosis. Attached Figure Description

[0044] Figure 1 A flowchart of the equipment fault diagnosis method provided in the embodiments of this specification;

[0045] Figure 2 A flowchart illustrating the equipment fault diagnosis method provided in the embodiments of this specification;

[0046] Figure 3 A flowchart illustrating the process of obtaining time-domain and frequency-domain features for the implementation of this specification;

[0047] Figure 4A flowchart illustrating the process of obtaining a device fault diagnosis model as provided in the embodiments of this specification;

[0048] Figure 5 A flowchart illustrating the process of obtaining a device fault diagnosis model as provided in the embodiments of this specification;

[0049] Figure 6 A schematic diagram illustrating the process of constructing a set of device waveform data samples provided for embodiments of this specification;

[0050] Figure 7 A schematic diagram of the equipment fault diagnosis device provided in the embodiments of this specification;

[0051] Figure 8 An internal structural diagram of a computer device provided for embodiments of this specification. Detailed Implementation

[0052] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0053] Taking fire-fighting equipment as an example, in the field of fire safety, there are some pain points and challenges in the methods of diagnosing fire-fighting equipment faults.

[0054] In related technologies, fault diagnosis techniques mainly rely on human (e.g., engineer) experience and simple threshold settings. This reliance has significant limitations when dealing with complex or hidden faults, often leading to misdiagnosis or missed diagnosis. For example, manual methods are influenced by subjective factors, resulting in inaccurate and unreliable diagnostic results.

[0055] Firefighting equipment generates a large amount of vibration data during operation, and this data is not only voluminous but also highly variable. Related technologies struggle to effectively process such high-dimensional and large-scale data. This insufficient processing capability leads to low efficiency in fault diagnosis and an inability to provide timely and accurate fault information.

[0056] In related technologies, some hidden or complex faults are often difficult to diagnose, requiring a lot of time and effort to locate and solve the problem.

[0057] In related technologies, firstly, first information and second information are acquired. Then, a fault knowledge graph is constructed based on the first information, including historical fault patterns and corresponding feature parameters. Next, a fault cause analysis is performed on the fault knowledge graph based on a pre-defined deep learning mathematical model to obtain fault analysis results. Based on the fault analysis results, multi-factor correlation assessment and risk prediction are conducted to establish a reliability prediction model. Finally, fault detection is performed on the railway information system based on the reliability prediction model and the second information to obtain real-time fault detection results. This implementation method, by combining a deep learning model and multi-factor correlation assessment, can more accurately identify and predict faults in railway information systems, effectively learn and identify complex fault patterns from a large amount of historical fault data, thereby improving the accuracy of fault detection.

[0058] However, the construction of knowledge graphs relies heavily on historical fault data and domain knowledge, resulting in significant limitations in capturing and adapting to newly emerging fault modes during the real-time operation of fire-fighting equipment. Furthermore, knowledge graph construction typically depends on manually defined features and rules, which may not fully and accurately reflect the complexity of fire-fighting equipment faults. In addition, because the construction of knowledge graphs is highly dependent on specific domain knowledge and historical data, their generalization ability may be limited. When faced with new and unseen fault modes, knowledge graphs may fail to provide effective diagnostic results.

[0059] Based on this, the embodiments of this specification provide a method for diagnosing equipment faults. First, target waveform data of the equipment to be tested is acquired. Next, feature extraction is performed on the target waveform data using an equipment fault diagnosis model to obtain time-domain and frequency-domain features. The equipment fault diagnosis model includes a semi-supervised learning module. This module utilizes unlabeled and labeled data to update the parameters of the equipment fault diagnosis model, adapting to dynamic changes in equipment status, effectively reducing dependence on data labeling, lowering training costs, and improving the model's generalization ability. Finally, the equipment fault diagnosis model performs equipment fault diagnosis based on the time-domain features, frequency-domain features, time-domain threshold range, and frequency-domain threshold range to obtain the equipment diagnosis result.

[0060] The above-described implementation reduces reliance on human experience through equipment fault diagnosis models, making the identification of equipment fault characteristics faster and more accurate. Furthermore, by achieving automated and intelligent fault diagnosis, the implementation improves the accuracy and efficiency of diagnostic results, reducing maintenance and repair costs. For vibration data under complex fault conditions, it can quickly locate hidden faults, improving the comprehensiveness and timeliness of fault diagnosis.

[0061] The equipment fault diagnosis method provided in this specification is applied to fire-fighting equipment. This equipment fault diagnosis method can integrate neural networks (CNN) and semi-supervised learning modules. For details, please refer to... Figure 1 Step 1: Sensor receives data: The initial equipment waveform data of the fire-fighting equipment is received by the sensor to form an initial equipment waveform data set.

[0062] Step 2: Data preprocessing: Clean and normalize the received initial device waveform data set to obtain the target waveform data set.

[0063] Step 3: Data Slicing: Divide the target waveform data into waveform segments corresponding to multiple appropriate time windows.

[0064] Step 4: Input data into the model for training: Input waveform segment data corresponding to multiple time windows into the CNN model and train it using a semi-supervised learning method.

[0065] Step 5: Model evaluation and tuning: Use the validation set to evaluate the trained model and adjust the model parameters to determine the best-performing equipment fault diagnosis model.

[0066] The above implementation combines sensor technology and deep learning models. By inputting the waveform data of fire-fighting equipment into the deep learning model for training and recognition, it is possible to discover potential patterns and rules in the data, and assist in quickly and accurately diagnosing the faults of fire-fighting equipment.

[0067] This specification provides a method for diagnosing equipment faults. Please refer to [link / reference]. Figure 2 Equipment fault diagnosis methods may include the following steps:

[0068] S210. Obtain the target waveform data of the device under test.

[0069] S220. The target waveform data is feature extracted using the equipment fault diagnosis model to obtain time-domain and frequency-domain features.

[0070] The equipment fault diagnosis model includes a feature semi-supervised learning module, which is used to update the parameters of the equipment fault diagnosis model.

[0071] Specifically, equipment generates various vibrations during operation. This vibration data not only reflects the equipment's operating status and performance but also provides rich information, serving as a crucial basis for fault diagnosis. Target waveform data of the equipment under test is acquired through sensors. Sensors can be mounted on the equipment itself or placed in appropriate locations nearby to capture vibration information. The target waveform data is then input into the equipment fault diagnosis model. The feature extraction module within the model extracts features from the input target waveform data, such as vibration frequency, vibration amplitude, and dynamic characteristics, yielding time-domain and frequency-domain features.

[0072] To enhance the generalization and adaptability of a model, the design of a semi-supervised learning module is crucial. Existing labeled data is used for supervised learning, calculating the loss and optimizing model parameters; while features extracted from unlabeled samples are used to generate pseudo-labels through self-learning or pseudo-labeling methods. These pseudo-labels are continuously updated during iterative model training, helping the model better adapt to constantly changing environments and scenarios, improving the accuracy of equipment fault diagnosis, and enhancing the model's generalization ability. For example, combining Generative Adversarial Networks (GANs) with semi-supervised learning and self-supervised learning can improve the model's performance and generalization ability.

[0073] In some implementations, data preprocessing can enhance data interpretability, thereby improving diagnostic results. For example, data cleaning and normalization can be performed on the waveform data corresponding to the device under test to obtain the target waveform data.

[0074] S230. The equipment fault diagnosis model performs equipment fault diagnosis based on time domain features, frequency domain features, time domain threshold range, and frequency domain threshold range, and obtains equipment diagnosis results.

[0075] Specifically, time-domain and frequency-domain features are important indicators for describing equipment vibration characteristics, providing rich information for subsequent fault diagnosis. Under normal equipment operation, the standard deviation of the time-domain features is calculated using historical data, and upper and lower limits are set as thresholds. In frequency-domain analysis, vibration frequencies are analyzed to determine the threshold range for specific frequencies; if these ranges are exceeded, the equipment is considered abnormal. The time-domain and frequency-domain threshold ranges ensure that faults can be identified promptly, avoiding potential production losses. The equipment fault diagnosis model can identify the presence of faults by comparing the extracted time-domain and frequency-domain features with the time-domain and frequency-domain threshold ranges. If either the time-domain or frequency-domain feature is outside the time-domain or frequency-domain threshold range, the fault diagnosis model will mark it as an abnormal state. By analyzing these differences and establishing the correspondence between features and faults, the equipment fault diagnosis model can accurately identify the fault type and obtain equipment diagnostic results, such as bearing damage, gear wear, or imbalance.

[0076] It should be noted that by comparing the characteristic differences between normal conditions and different failure modes, a correspondence between characteristics and faults can be established, thereby enabling accurate identification and diagnosis of fire equipment faults. This is crucial for timely detection and handling of potential problems in fire equipment, ensuring its normal operation and safety.

[0077] Furthermore, deep learning methods such as Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), or attention mechanisms can be used when selecting equipment fault diagnosis models. These methods can also be combined with traditional signal processing techniques to improve the accuracy of fault diagnosis. Specifically, the feature extraction module can employ adaptive filters and time-frequency analysis methods (such as Short-Time Fourier Transform (STFT) and wavelet transform).

[0078] In the above implementation, firstly, target waveform data of the device to be tested is acquired. Next, feature extraction is performed on the target waveform data using a device fault diagnosis model to obtain time-domain and frequency-domain features. The device fault diagnosis model includes a feature extraction module and a semi-supervised learning module. The semi-supervised learning module utilizes unlabeled and labeled data to update the parameters of the device fault diagnosis model, adapting to dynamic changes in the device's state, effectively reducing dependence on data labeling, lowering training costs, and improving the model's generalization ability. Finally, the device fault diagnosis model performs device fault diagnosis based on time-domain features, frequency-domain features, time-domain threshold range, and frequency-domain threshold range to obtain the device diagnosis result.

[0079] The above-described implementation reduces reliance on human experience through equipment fault diagnosis models, making the identification of equipment fault characteristics faster and more accurate. Furthermore, by achieving automated and intelligent fault diagnosis, the implementation improves the accuracy and efficiency of diagnostic results, reducing maintenance and repair costs. For vibration data under complex fault conditions, it can quickly locate hidden faults, improving the comprehensiveness and timeliness of fault diagnosis.

[0080] In some implementations, please refer to Figure 3 The process of extracting features from target waveform data using a device fault diagnosis model to obtain time-domain and frequency-domain features may include the following steps:

[0081] S310. Divide the target waveform data into waveform segment data corresponding to multiple time windows.

[0082] Specifically, a suitable time window length (e.g., 1 second, 2 seconds, etc.) should be selected, and this length should be determined based on the characteristics of the data and the analysis requirements. Several factors need to be considered when selecting the time window length, such as the data sampling frequency, the rate of signal change, and the analysis objective (e.g., feature extraction, classification, or prediction). For signals with high-frequency dynamic changes, a shorter time window may be needed to ensure that all important features are captured. For relatively stable signals, a longer time window can be selected to reduce the complexity of data processing. After setting the time window, starting from the initial point, the target waveform data is sequentially divided into waveform segments corresponding to multiple time windows using a sliding window technique.

[0083] S320. The vibration frequency characteristics of waveform segments corresponding to multiple time windows are extracted using the equipment fault diagnosis model to obtain frequency domain characteristics.

[0084] Specifically, equipment generates various vibrations during operation. This vibration data contains rich information, reflecting the equipment's operational status and performance, and is a crucial basis for fault diagnosis. Waveform segments corresponding to multiple time windows are used as input to the equipment fault diagnosis model. The model extracts vibration frequency characteristics from these waveform segments, refining useful information to obtain frequency domain features. These frequency domain features reveal the distribution of vibration energy generated during equipment operation, which is very helpful in identifying different operating states and potential faults. Identifying these features is crucial for determining the equipment's operating status and potential faults. In frequency domain feature analysis, vibration frequency is used to identify equipment fault modes; for example, looseness may lead to an increase in frequency components.

[0085] S330. By using the equipment fault diagnosis model, the vibration amplitude characteristics of waveform segments corresponding to multiple time windows are extracted to obtain time-domain characteristics.

[0086] Specifically, waveform segment data corresponding to multiple time windows are used as input to the equipment fault diagnosis model. The model extracts vibration amplitude characteristics from these waveform segment data, refining useful information for analysis and obtaining time-domain features. These time-domain features describe the variation of vibration signals over time, reflecting the equipment's stability, wear level, and the presence of abnormal vibrations. For example, if the time-domain features indicate an abnormally high vibration amplitude, it may suggest internal loosening of the equipment.

[0087] In the above embodiments, the target waveform data is divided into waveform segments corresponding to multiple time windows. The vibration frequency features of the waveform segments corresponding to multiple time windows are extracted by the equipment fault diagnosis model to obtain frequency domain features. The vibration amplitude features of the waveform segments corresponding to multiple time windows are extracted by the equipment fault diagnosis model to obtain time domain features, thereby improving the accuracy of fault diagnosis.

[0088] In some implementations, please refer to Figure 4 The equipment fault diagnosis model is obtained through the following methods:

[0089] S410, Construct a set of device waveform data samples.

[0090] S420. Based on the semi-supervised learning module, the parameters of the fault diagnosis model of the equipment to be trained are updated using the set of equipment waveform data samples to obtain the equipment fault diagnosis model.

[0091] Specifically, device-related waveform data can be collected from multiple channels such as public datasets and sensor data. A portion of the collected waveform data is then labeled to construct a device waveform data sample set. To fully utilize unlabeled data, a semi-supervised learning module is introduced. This module can update model parameters using both labeled and unlabeled device waveform data samples during training. By using pseudo-labels and contrastive learning techniques, a special loss function is designed, enabling the model to learn not only supervised learning on labeled data but also important information from unlabeled data.

[0092] The supervised learning component uses labeled data to guide the model in learning the distinguishing features between different failure modes. The unsupervised learning component, on the other hand, helps the model discover inherent patterns and regularities within the data. To fully utilize both labeled and unlabeled data, the semi-supervised learning module cleverly combines the advantages of both methods, aiming to leverage both labeled and large amounts of unlabeled data to train the model and enhance its diagnostic capabilities.

[0093] For example, fire-fighting equipment typically operates in complex and variable environments, such as high temperatures, humidity variations, and different mechanical loads. These environmental factors cause the vibration data collected by sensors to be affected by various noises and interferences. Specifically, mechanical vibrations may originate from friction, collisions, etc., during equipment operation, while electromagnetic interference may be caused by the operation of nearby electrical equipment. Furthermore, multiple vibration sources may exist within the equipment, such as the simultaneous operation of motors and pumps. These factors interact, greatly increasing the complexity and uncertainty of the data. Moreover, due to the special and diverse nature of fire-fighting equipment, acquiring labeled data is usually costly and time-consuming. For example, acquiring labeled data requires long-term monitoring and analysis by professionals to accurately identify fault modes. However, in practical applications, a large amount of unlabeled vibration data can usually be easily acquired, such as daily operating records of the equipment and real-time sensor monitoring. In contrast, labeled data is relatively scarce and expensive.

[0094] In this context, semi-supervised learning methods are particularly suitable. Semi-supervised learning can simultaneously utilize limited labeled data and abundant unlabeled data, learning explicit features of fault modes from labeled data while simultaneously mining hidden vibrational patterns and potential structural information from unlabeled data. In this way, even with limited labeled data, the model's generalization ability and diagnostic performance can be improved, enabling it to exhibit good diagnostic performance when faced with new and unseen fault data.

[0095] In the above implementation, a set of device waveform data samples is constructed. Based on the semi-supervised learning module, the parameters of the device fault diagnosis model to be trained are updated using the set of device waveform data samples to obtain the device fault diagnosis model, thereby improving the generalization ability and training efficiency of the model.

[0096] In some implementations, please refer to Figure 5 The device waveform data sample set includes both unlabeled and labeled waveform data sample sets. Based on the semi-supervised learning module, the parameters of the device fault diagnosis model to be trained are updated using the device waveform data sample set to obtain the device fault diagnosis model. This may include the following steps:

[0097] S510. Use the unlabeled waveform data sample set to pre-train the fault diagnosis model of the equipment to be trained, update the parameters of the fault diagnosis model of the equipment to be trained, and obtain the intermediate equipment fault diagnosis model.

[0098] Specifically, to enable the equipment fault diagnosis model to learn general features, it is first necessary to pre-train the model using a large-scale set of unlabeled waveform data samples. This provides a better starting point for subsequent specific fault diagnosis tasks. The unlabeled waveform data sample set contains waveform data under various fault conditions, ensuring that the model can learn rich features. During training, the model updates its parameters by minimizing the prediction loss. This means the model continuously adjusts its weights to reduce the difference between the predicted and actual results. Commonly used loss functions include cross-entropy loss or regression loss, depending on the task requirements. After pre-training, the model's parameters are saved; this model is then called an intermediate equipment fault diagnosis model, which has learned some useful features and patterns from the unlabeled data.

[0099] S520. Use the labeled waveform data sample set to train the intermediate equipment fault diagnosis model, update the parameters of the intermediate equipment fault diagnosis model, and obtain the equipment fault diagnosis model.

[0100] Specifically, the intermediate equipment fault diagnosis model can be further fine-tuned to adapt to labeled waveform datasets, making the model better suited for equipment fault diagnosis tasks and thus enhancing its performance. The intermediate equipment fault diagnosis model is trained using a set of labeled waveform data samples. During training, the difference between the model output and the true labels is measured by calculating the prediction loss (e.g., cross-entropy loss or mean squared error). An optimization algorithm is used to update the parameters of the intermediate equipment fault diagnosis model, aiming to minimize this loss function, thereby improving the model's accuracy and reliability, resulting in the final equipment fault diagnosis model.

[0101] In some implementations, a validation set is used to evaluate the performance of the equipment fault diagnosis model on equipment fault diagnosis tasks. Evaluation metrics include precision, recall, and F1 score, which reflect the model's diagnostic performance from different perspectives. Precision measures the model's ability to correctly identify fault samples, while recall reflects the model's ability to identify all actual fault samples. The F1 score, the harmonic mean of precision and recall, provides a comprehensive indicator of model performance. Furthermore, AUC-ROC curves can be used to visually demonstrate the model's diagnostic performance at different thresholds. The closer the AUC value is to 1, the better the model's performance. AUC-ROC curves provide a more comprehensive understanding of the model's performance.

[0102] The validation set is used to select qualified models for evaluation on the test set. If a model's evaluation metrics on the test set do not meet the preset requirements, a new model selection process will be conducted, potentially involving hyperparameter tuning and model structure adjustments to ensure improved model performance and reliability and stability in real-world applications. For example, model structure tuning may involve adding or removing layers, adjusting layer parameters, etc., to increase model complexity and learning capacity. Hyperparameter tuning may include optimizing the learning rate, batch size, etc.

[0103] The ultimate goal of tuning is to improve the model's generalization ability and diagnostic accuracy, ensuring that the model exhibits stable diagnostic performance even when faced with new and unseen fault data. To achieve this goal, multiple rounds of evaluation and tuning will be conducted, iteratively optimizing the model until a satisfactory performance level is reached.

[0104] Through continuous evaluation and optimization, we ensure that the trained model can effectively identify equipment malfunctions in practical applications, providing accurate and reliable support for equipment maintenance and troubleshooting.

[0105] In the above implementation, the fault diagnosis model of the device to be trained is pre-trained using an unlabeled waveform data sample set, and the parameters of the fault diagnosis model are updated to obtain an intermediate device fault diagnosis model. Then, the intermediate device fault diagnosis model is trained using a labeled waveform data sample set, and the parameters of the intermediate device fault diagnosis model are updated to obtain a device fault diagnosis model. This enables rapid and accurate fault identification and location, providing strong support for equipment maintenance and fault diagnosis.

[0106] In some implementations, please refer to Figure 6 Constructing a set of device waveform data samples may include the following steps:

[0107] S610: Use sensors to acquire an initial set of device waveform data.

[0108] Specifically, sensors are used to monitor equipment and acquire an initial set of equipment waveform data related to its operating status. For example, sensors can be used to acquire the initial set of waveform data for fire-fighting equipment. Sensors should be placed in optimal locations based on the equipment's operating environment and characteristics to capture accurate equipment information. For instance, sensors can be placed inside or near the equipment at appropriate locations. The initial set of equipment waveform data can include various equipment fault scenarios, including but not limited to normal operation, minor wear, and severe faults, ensuring data diversity and comprehensiveness. This diverse dataset helps improve the generalization ability of subsequent fault diagnosis models, enhancing their adaptability and accuracy in practical applications. The waveform data can include information on vibration amplitude and frequency.

[0109] S620. Perform data cleaning and normalization on the initial device waveform data set to obtain the target waveform data set.

[0110] Specifically, data cleaning is a crucial step in ensuring data quality. The initial set of device waveform data undergoes data cleaning, including noise removal, data smoothing, and outlier elimination, to ensure data accuracy and consistency. After cleaning, the cleaned waveform data is normalized to convert data of different scales to the same scale, preventing overfitting or underfitting during model training due to data scaling issues, thereby improving algorithm efficiency and accuracy.

[0111] In some implementations, noise removal can be achieved using adaptive filtering. Adaptive filtering is highly adaptable, adjusting filter parameters in real time based on the characteristics of the initial device waveform data set to accommodate changes in the initial device waveform data set. Compared to fixed filters, it offers greater adaptability and better noise reduction performance. Adaptive filtering can dynamically adjust filter parameters according to the characteristics of the initial device waveform data set, more effectively removing noise while preserving the valid information in the initial device waveform data set.

[0112] For example, the Least Mean Square (LMS) filter is an adaptive filtering technique designed to remove noise and preserve valid signals by adjusting filter parameters in real time. Its working mechanism is based on minimizing the mean square error between the output signal and the desired signal. Specifically, filter parameters are first initialized, and then an initial set of device waveform data is input for filtering. During the filtering process, parameters are adjusted in real time to adapt to data changes. By calibrating filter parameters in real time, the adaptive filter can dynamically remove noise and preserve valid signal information, improving data quality. This method can effectively address data denoising needs under various data characteristics and complex environments.

[0113] Normalization can be achieved using the min-max normalization method, as shown in the following formula:

[0114]

[0115] Where, x norm x is the target waveform data, and x is the cleaned waveform data. min It is the minimum value in the cleaned waveform data, x max It is the minimum value in the cleaned waveform data.

[0116] S630. Data annotation is performed on a portion of the target waveform data in the target waveform data set to obtain a device waveform data sample set.

[0117] Specifically, a subset of target waveform data is selected from the target waveform dataset for data annotation. Annotation includes labeling time-domain features (such as vibration amplitude) and frequency-domain features (such as vibration frequency). Specialized annotation tools can be used to ensure the accuracy and consistency of the annotations. The annotated target waveform data will be used in the supervised learning phase of model training to help the model learn the frequency and amplitude characteristics of the waveform data. The annotated and unannotated target waveform data together constitute the device waveform data sample set.

[0118] The device waveform data sample set is divided into training, validation, and test sets to ensure the model's generalization ability on different data. The typical division ratio is 70%, 15%, and 15%.

[0119] In the above embodiments, an initial set of device waveform data is acquired using sensors, the initial set of device waveform data is cleaned and normalized to obtain a target set of waveform data, and a portion of the target waveform data in the target set is labeled to obtain a set of device waveform data samples, so as to improve data quality and reliability, thereby improving the accuracy of the model.

[0120] In some implementations, the equipment fault diagnosis model is built based on the convolutional neural network (CNN) model.

[0121] Specifically, equipment fault diagnosis is crucial for ensuring equipment safety and reliability. To improve equipment fault identification capabilities, equipment fault diagnosis models can be built based on Convolutional Neural Network (CNN) models. CNN models consist of multiple layers, including convolutional layers, pooling layers, and fully connected layers, to progressively extract high-level abstract features from waveform data, such as frequency domain features and time domain features. It should be noted that CNN models trained on large-scale datasets have stronger generalization capabilities and can better handle unknown situations.

[0122] During model training, CNNs gradually learn the unique characteristics of different failure modes. These characteristics may be closely related to specific vibration frequencies, amplitudes, and waveform shapes. For example, bearing failures may manifest as increased vibrations at specific frequencies.

[0123] In the above embodiments, the equipment fault diagnosis model is built based on a Convolutional Neural Network (CNN) model. CNNs can automatically extract high-level features, which are typically richer and more discriminative. CNNs can accurately identify different types of fault features, reducing reliance on human experience and thus improving the accuracy of diagnostic results. This means that fault types can be identified more accurately and reliably, providing crucial protection for equipment safety and reliability.

[0124] This specification provides a model training method for training a device fault diagnosis model based on any one of the above.

[0125] Specifically, an initial set of device waveform data is acquired using sensors. This initial set of waveform data is then cleaned and normalized to obtain a target set of waveform data. A portion of the target waveform data is then labeled to obtain a set of device waveform data samples. This set of sample waveform data includes both unlabeled and labeled sets. The unlabeled set of sample waveform data is used to pre-train the device fault diagnosis model, updating its parameters to obtain an intermediate device fault diagnosis model. The labeled set of sample waveform data is then used to train the intermediate model, updating its parameters to obtain the final device fault diagnosis model. This final device fault diagnosis model is built upon a Convolutional Neural Network (CNN) model.

[0126] This specification provides an embodiment of a device fault diagnosis apparatus 700. Please refer to [link / reference]. Figure 7 The equipment fault diagnosis device 700 includes: a waveform data acquisition module 710, a time-frequency domain feature extraction module 720, and an equipment fault diagnosis module 730.

[0127] The waveform data acquisition module 710 is used to acquire the target waveform data of the device under test.

[0128] The time-frequency domain feature extraction module 720 is used to extract features from the target waveform data through the equipment fault diagnosis model to obtain time-domain features and frequency-domain features. The equipment fault diagnosis model includes a semi-supervised learning module, which is used to update the parameters of the equipment fault diagnosis model.

[0129] The equipment fault diagnosis module 730 is used to perform equipment fault diagnosis based on the time domain features, the frequency domain features, the time domain threshold range, and the frequency domain threshold range, and obtain equipment diagnosis results.

[0130] For a detailed description of the equipment fault diagnosis device, please refer to the description of the equipment fault diagnosis method above, which will not be repeated here.

[0131] In some embodiments, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a device fault diagnosis method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0132] Those skilled in the art will understand that Figure 8 The structures shown are merely block diagrams of some structures related to the solutions disclosed in this specification, and do not constitute a limitation on the computer device to which the solutions disclosed in this specification are applied. Specifically, the computer device may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements.

[0133] In some embodiments, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method steps described above.

[0134] This specification provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method in any of the above embodiments.

[0135] One embodiment of this specification provides a computer program product including instructions that, when executed by a processor of a computer device, enable the computer device to perform the steps of the method described in any of the above embodiments.

[0136] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

Claims

1. A method for diagnosing equipment faults, characterized in that, The method includes: Acquire the target waveform data of the device under test; The target waveform data is subjected to feature extraction by the equipment fault diagnosis model to obtain time domain features and frequency domain features. The equipment fault diagnosis model includes a semi-supervised learning module, which is used to update the parameters of the equipment fault diagnosis model. The equipment fault diagnosis model performs equipment fault diagnosis based on the time-domain features, the frequency-domain features, the time-domain threshold range, and the frequency-domain threshold range, and obtains the equipment diagnosis results.

2. The method according to claim 1, characterized in that, The step of extracting features from the target waveform data using a device fault diagnosis model to obtain time-domain and frequency-domain features includes: The target waveform data is divided into waveform segment data corresponding to multiple time windows; The vibration frequency features are extracted from the waveform segment data corresponding to the multiple time windows using the equipment fault diagnosis model to obtain the frequency domain features. The vibration amplitude features are extracted from the waveform segment data corresponding to the multiple time windows using the equipment fault diagnosis model to obtain the time domain features.

3. The method according to claim 1, characterized in that, The equipment fault diagnosis model is obtained through the following method: Construct a set of device waveform data samples; Based on the semi-supervised learning module, the parameters of the device fault diagnosis model to be trained are updated using the device waveform data sample set, thereby obtaining the device fault diagnosis model.

4. The method according to claim 3, characterized in that, The device waveform data sample set includes an unlabeled waveform data sample set and a labeled waveform data sample set; the semi-supervised learning module uses the device waveform data sample set to update the parameters of the device fault diagnosis model to be trained, thereby obtaining the device fault diagnosis model, including: The fault diagnosis model of the device to be trained is pre-trained using the unlabeled waveform data sample set, and the parameters of the fault diagnosis model of the device to be trained are updated to obtain an intermediate device fault diagnosis model. The intermediate equipment fault diagnosis model is trained using the labeled waveform data sample set, and the parameters of the intermediate equipment fault diagnosis model are updated to obtain the equipment fault diagnosis model.

5. The method according to claim 3, characterized in that, The constructed device waveform data sample set includes: Use sensors to acquire an initial set of device waveform data; The initial device waveform data set is cleaned and normalized to obtain the target waveform data set; Data annotation is performed on a portion of the target waveform data in the target waveform data set to obtain the device waveform data sample set.

6. The method according to any one of claims 1 to 5, characterized in that, The equipment fault diagnosis model is built based on the convolutional neural network (CNN) model.

7. A model training method, characterized in that, The model training method is used to train the equipment fault diagnosis model according to any one of claims 1-6, and the model training method includes: Construct a set of device waveform data samples; Based on the semi-supervised learning module, the parameters of the device fault diagnosis model to be trained are updated using the device waveform data sample set, thereby obtaining the device fault diagnosis model.

8. The method according to claim 7, characterized in that, The device waveform data sample set includes an unlabeled waveform data sample set and a labeled waveform data sample set; the semi-supervised learning module uses the device waveform data sample set to update the parameters of the device fault diagnosis model to be trained, thereby obtaining the device fault diagnosis model, including: The fault diagnosis model of the device to be trained is pre-trained using the unlabeled waveform data sample set, and the parameters of the fault diagnosis model of the device to be trained are updated to obtain an intermediate device fault diagnosis model. The intermediate equipment fault diagnosis model is trained using the labeled waveform data sample set, and the parameters of the intermediate equipment fault diagnosis model are updated to obtain the equipment fault diagnosis model.

9. The method according to claim 7, characterized in that, The constructed device waveform data sample set includes: Use sensors to acquire an initial set of device waveform data; The initial device waveform data set is cleaned and normalized to obtain the target waveform data set; Data annotation is performed on a portion of the target waveform data in the target waveform data set to obtain the device waveform data sample set.

10. The method according to any one of claims 7 to 9, characterized in that, The equipment fault diagnosis model is built based on the convolutional neural network (CNN) model.

11. A device for diagnosing equipment faults, characterized in that, The device includes: The waveform data acquisition module is used to acquire the target waveform data of the device under test; The time-frequency domain feature extraction module is used to extract features from the target waveform data through the equipment fault diagnosis model to obtain time-domain features and frequency-domain features. The equipment fault diagnosis model includes a semi-supervised learning module, which is used to update the parameters of the equipment fault diagnosis model. The equipment fault diagnosis module is used to perform equipment fault diagnosis based on the time domain features, the frequency domain features, the time domain threshold range, and the frequency domain threshold range, and obtain equipment diagnosis results.

12. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 10.

Images