Electromechanical equipment abnormity identification and fault prediction method based on multi-mode OOD detection

Abstract

The invention discloses an electromechanical equipment abnormity identification and fault prediction method based on multi-modal OOD detection, and belongs to the field of electromechanical equipment detection, and the method comprises the steps: S1, collecting multi-modal sensor time sequence data of an electromechanical equipment cluster, and representing the multi-modal sensor time sequence data as a three-dimensional data structure; the time sequence data of the multi-mode sensor are ID data which are not abnormal; s2, training the AE network by using the ID data, and freezing parameters of the trained AE network; s3, on the basis of the AE network of the frozen parameters, constructing an OOD detector; and S4, performing anomaly identification on the input sample of the three-dimensional data structure by using an OOD detector, and performing abnormal equipment positioning based on the reconstruction loss value of each modal sensor on the equipment dimension to realize anomaly identification and fault prediction. Through the multi-modal data fusion and OOD detection technology, the limitation of a traditional method is overcome, and the fault diagnosis effect which is more accurate, better in robustness and capable of being explained is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of electromechanical equipment detection, and specifically relates to a method for electromechanical equipment anomaly identification and fault prediction based on multimodal OOD detection. Background Art

[0002] Currently, in the field of anomaly identification and fault prediction of electromechanical equipment, traditional methods have two key technical bottlenecks: one is the lack of explainability of fault location, and the other is the high rate of missed detection of unknown faults.

[0003] Although existing technologies can perform fault detection based on single-modal data such as vibration, current, and temperature, due to the lack of an effective multi-modal feature fusion mechanism and an explainable analysis framework, the system can only output simple abnormality alarms and cannot clearly indicate the specific component where the fault occurred and its cause, which brings great trouble to subsequent maintenance decisions.

[0004] More seriously, current mainstream methods generally adopt a closed-set learning paradigm, assuming that all fault types have already occurred during the training phase. This makes the system ineffective when faced with new fault modes that may arise in actual operation, significantly increasing the missed detection rate. This limitation is particularly prominent in industrial sites, because the failure modes of electromechanical equipment are dynamic and affected by multiple factors such as the environment and load. Methods that rely solely on historical data for supervised learning are unable to meet the fault detection needs in such open environments.

[0005] Although deep learning technology has shown advantages in automatic feature extraction in recent years, most models still remain at the "black box" operation level, unable to explain the physical mechanism of fault occurrence, nor effectively identify abnormal patterns outside the distribution of training data, which seriously restricts the promotion and application of intelligent diagnosis systems in actual industrial scenarios.

[0006] Currently, complex electromechanical systems contain a wide variety of equipment, complex operating states, and diverse fault patterns. Traditional fault diagnosis methods struggle to meet the requirements of high accuracy, strong robustness, and good interpretability. First, traditional supervised learning methods based on the closed set assumption can typically only identify a limited number of fault modes known during training. Their recognition ability for out-of-distribution (OOD) samples—unknown faults that did not appear in the training set—is weak, resulting in a high rate of missed detection and a serious safety hazard. Second, most existing systems can only issue simple anomaly alarms based on anomaly thresholds or predicted probabilities, but are unable to further locate the specific device or component that caused the fault, let alone provide an effective explanation of the fault's mechanism or characteristics. This limits their usability and reliability in real-world industrial scenarios. Summary of the Invention

[0007] In response to the above-mentioned deficiencies in the prior art, the method for anomaly identification and fault prediction of electromechanical equipment based on multimodal OOD detection provided by the present invention solves the problems of lack of explainability in fault location and high missed detection rate of unknown faults when traditional methods are used to implement anomaly identification and fault prediction of electromechanical equipment.

[0008] In order to achieve the above-mentioned purpose of the invention, the technical solution adopted by the present invention is: a method for abnormality identification and fault prediction of electromechanical equipment based on multimodal OOD detection, comprising the following steps: S1, collects multimodal sensor time series data of the electromechanical equipment cluster and represents it as a three-dimensional data structure; The multimodal sensor time series data is ID data without any abnormality; S2. Use ID data to train the AE network and freeze the trained AE network parameters; S3. Based on the AE network with frozen parameters, an OOD detector is constructed; S4. Use the OOD detector to identify anomalies in the input samples of the three-dimensional data structure, and locate abnormal devices based on the reconstruction loss value of each modal sensor in the device dimension to achieve anomaly identification and fault prediction.

[0009] Furthermore, in step S1, the three-dimensional data structure of the sensor time series data is represented as (N, D, S), where N is the number of samples, indicating the number of observations collected at a set time or time period, D is the number of electromechanical devices, indicating the total number of independent devices in the electromechanical device cluster, and S indicates the number of sensor types configured for each device.

[0010] Furthermore, in step S2, the AE network includes an encoder and a decoder that maintain a symmetric structure; The method for training the AE network is specifically as follows: The input data is mapped to a latent space by an encoder, and the latent feature data is mapped back to the original space by a decoder to obtain reconstructed data; wherein the latent feature data in the latent space is fused with multimodal feature information of the ID data; The reconstruction error between the input ID data and the reconstructed data is used as the loss function to train the AE network.

[0011] Furthermore, the process of mapping the input data to the latent space through the encoder is as follows: Perform linear transformation on the input data and convert it into latent space data; According to the attention function learned from the latent space data, a weight is assigned to each modality of the input data to obtain its corresponding assignable weight; The normalized attention weight is obtained by softmax operation on the assignable weight; Use attention weights to perform weighted summation on latent space data to obtain fused feature data that fuses multimodal feature information; The fused feature data is mapped to the latent space using a fully connected layer to obtain the latent feature data.

[0012] Furthermore, the potential feature data Expressed as: Where, represents the fusion feature data, represents the weights for mapping fused feature data to the latent space, represents the third bias term, represents the relu activation function, represents the normalized attention weight of the kth moment in the dth sensor in the nth device, represents the assignable weight of the dth sensor in the nth device, represents the allocable weight of the kth sensor in the nth device at time k, represents the assignable weight of the jth sensor in the nth device at time j, represents latent space data, represents the latent space data corresponding to the kth moment in the dth sensor in the nth device, represents the second layer projection vector, represents a nonlinear activation function, represents the second bias term, represents the first layer linear transformation weight, represents the second bias term, Represents the input data, which represents the sensor observation vector under each device n and sensor d. The superscript T represents the transpose operation, and the subscripts j and k are both time ordinal numbers. .

[0013] Furthermore, in step S2, the loss function of the AE network is trained for: Where, represents the input data of the dth sensor in the nth device at time s, represents the reconstructed data of the d-th sensor in the n-th device at time s, represents the number of samples, Represents the number of electromechanical devices, and S represents the number of sensor types configured for each device.

[0014] Furthermore, in step S3, the OOD detector is used to evaluate the degree of deviation of the input sample relative to the ID data distribution, thereby achieving OOD data recognition; The method for the OOD detector to perform OOD data recognition is: Calculate the reconstruction loss value of the input sample after passing through the parameter-frozen AE network, and compare the reconstruction loss value with the set discrimination threshold. When the reconstruction loss value exceeds the set discrimination threshold, the input sample is OOD data, otherwise it is ID data; The OOD data is abnormal data.

[0015] Furthermore, the method for determining the discrimination threshold is specifically as follows: For the training set containing only ID data, calculate the reconstruction loss value after passing through the AE network with frozen parameters; Count the mean of the reconstruction loss values ​​of all ID data and standard deviation , and then determine the discrimination threshold .

[0016] Furthermore, in step S4, the method for locating the abnormal device is specifically as follows: For input samples identified as abnormal, the reconstruction loss values ​​of all sensor channels on each device are accumulated item by item, and then the total reconstruction loss value in the device dimension is calculated. The device with the largest total reconstruction loss value is regarded as the abnormal device to be located.

[0017] Furthermore, the total reconstruction loss value for: Where, represents the input data of the dth sensor in the nth device at time s, Represents the reconstructed data of the d-th sensor in the n-th device at time s.

[0018] The beneficial effects of the present invention are: This invention overcomes the limitations of traditional methods by combining multimodal data fusion and OOD detection technology, and brings more accurate, robust and explainable fault diagnosis results, which are specifically reflected in: (1) Unlike traditional methods that rely on single modal data, the present invention integrates multimodal data such as vibration, temperature, and current and represents them in the form of a three-dimensional data structure. Through this multi-dimensional time series structure, not only can the operating status of a single device be tracked, but also the correlation pattern and fault propagation path between devices in the cluster can be analyzed from a global perspective, thereby enhancing the stability and accuracy of the system in the absence of data or noise interference, and providing a solid data foundation for subsequent intelligent operation and maintenance, predictive maintenance, and risk warning.

[0019] (2) The present invention adopts the unsupervised learning method of AE network to effectively identify unknown fault modes, and adopts the data fusion method based on modal attention mechanism to fuse multiple sensor data. The discrimination threshold is set based on the reconstruction loss value, which can accurately distinguish normal data from abnormal data, thereby improving the system's adaptability to new faults.

[0020] (3) In terms of fault location, the present invention accurately locates the faulty device by analyzing the device-level reconstruction error, and can effectively fuse the abnormal signals at the modal level, thereby enhancing the robustness and reliability of abnormality location and avoiding the global alarm problem of traditional methods.

[0021] (4) The present invention automatically sets a reasonable discrimination threshold through statistical methods. This method does not rely on any out-of-distribution data or label information, and is therefore particularly suitable for unsupervised or semi-supervised scenarios. On the other hand, the threshold mechanism constructed through statistics also enhances the stability and robustness of the detection process because it utilizes the quantitative relationship between the central tendency of the distribution and the degree of outliers, thereby being able to effectively distinguish normal samples from potential abnormal samples, and can maintain consistent detection results in different data sets or application scenarios, reducing false alarms.

[0022] In general, the present invention demonstrates higher accuracy and robustness in fault detection and location, providing strong support for equipment maintenance and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Flowchart of the method for anomaly identification and fault prediction of electromechanical equipment based on multimodal OOD detection provided by the present invention. DETAILED DESCRIPTION

[0024] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0025] The present invention proposes a method for anomaly identification and fault prediction that integrates multimodal information and OOD detection mechanism to achieve comprehensive perception of the operating status of electromechanical equipment, accurate identification of faults, and interpretable positioning of their locations and causes. Specifically, the method for anomaly identification and fault prediction of electromechanical equipment based on multimodal OOD detection in the present invention is as follows: Figure 1 As shown, the following steps are included: S1, collects multimodal sensor time series data of the electromechanical equipment cluster and represents it as a three-dimensional data structure; The multimodal sensor time series data is ID data without any abnormality; S2. Use ID data to train the AE network and freeze the trained AE network parameters; S3. Based on the AE network with frozen parameters, an OOD detector is constructed; S4. Use the OOD detector to identify anomalies in the input samples of the three-dimensional data structure, and locate abnormal devices based on the reconstruction loss value of each modal sensor in the device dimension to achieve anomaly identification and fault prediction.

[0026] In step S1 of an embodiment of the present invention, the three-dimensional data structure of the sensor time series data is represented as (N, D, S), where N is the number of samples, indicating the number of observations collected at a set time or time period, D is the number of electromechanical devices, indicating the total number of independent devices in the electromechanical device cluster, and S indicates the number of sensor types configured for each device.

[0027] In this embodiment, a multi-dimensional temporal organization method for multi-electromechanical device clusters is proposed. This method effectively expresses and manages the complex data structure of multiple devices, multiple sensors, and multiple time points in large-scale electromechanical systems, and represents it as a three-dimensional data structure (N, D, S), where N represents the number of samples, specifically the number of observations collected at a certain moment or within a certain time period; D represents the number of electromechanical devices, that is, the total number of independent devices included in the cluster; and S represents the number of sensor types configured for each device, such as common temperature, vibration, current, and other types of sensors.

[0028] Specifically, this three-dimensional data structure has good scalability and organization, and can clearly capture data information at different levels of the system (time, equipment, sensors), enabling structured expression and efficient access to data. Compared to traditional two-dimensional tables or flat storage methods, the use of a three-dimensional structure not only reduces data redundancy and improves storage and processing efficiency, but also facilitates subsequent applications in tasks such as machine learning, condition monitoring, and anomaly detection. For example, when using deep learning models for modeling and analysis, this data structure is naturally adapted to multi-dimensional tensor inputs, fully preserving the correlation of the original data in time, space, and functional dimensions, thereby enhancing the model's understanding and prediction capabilities of system behavior.

[0029] Furthermore, in complex industrial environments where multiple devices operate collaboratively, dynamic coupling or mutual influence may exist between different devices. This three-dimensional data structure not only tracks the operating status of individual devices but also analyzes the correlation patterns and fault propagation paths between devices in the cluster from a global perspective, providing a solid data foundation for subsequent intelligent operations, predictive maintenance, and risk warnings.

[0030] In step S2 of this embodiment, to effectively integrate multimodal data collected by various sensors (such as temperature, vibration, and current) in electromechanical equipment, the present invention proposes an AE network that incorporates a modal attention mechanism. This network models the ID data of the normal operating state (in-distribution, ID) and accurately detects OOD data (out-of-distribution data) through reconstruction error. When the reconstruction error of a test sample significantly deviates from the training distribution, it is identified as a possible unknown fault.

[0031] The AE network in the embodiment of the present invention includes an encoder and a decoder that maintain a symmetric structure; The method for training the AE network is specifically as follows: The input data is mapped to a latent space by an encoder, and the latent feature data is mapped back to the original space by a decoder to obtain reconstructed data; wherein the latent feature data in the latent space is fused with multimodal feature information of the ID data; The reconstruction error between the input ID data and the reconstructed data is used as the loss function to train the AE network.

[0032] Specifically, based on the above AE network structure, the process of mapping the input data to the latent space through the encoder is as follows: Perform linear transformation on the input data and convert it into latent space data; According to the attention function learned from the latent space data, a weight is assigned to each modality of the input data to obtain its corresponding assignable weight; The normalized attention weight is obtained by softmax operation on the assignable weight; Use attention weights to perform weighted summation on latent space data to obtain fused feature data that fuses multimodal feature information; The fused feature data is mapped to the latent space using a fully connected layer to obtain the latent feature data.

[0033] Specifically, under each sensor d of device n, the sensor observation vector , perform linear transformation on the original data to convert it into more discriminative latent space data According to the latent space data Computational learning of attention functions Assign weights to each modality, you can assign weights for: Then apply the softmax operation to the assignable weights to obtain the normalized attention weights for: Using attention weights Perform weighted summation on the latent space data to obtain fused feature data for: Through the above-mentioned modal attention mechanism, the AE network can adaptively capture the importance of different sensors under specific time and device conditions, thereby generating more discriminative fusion representations. The fused feature data is mapped to the latent space using a fully connected layer to obtain the latent feature data for: Where, represents the fusion feature data, represents the weights for mapping fused feature data to the latent space, represents the third bias term, represents the relu activation function, represents the normalized attention weight of the kth moment in the dth sensor in the nth device, represents the assignable weight of the dth sensor in the nth device, represents the allocable weight of the kth sensor in the nth device at time k, represents the assignable weight of the jth sensor in the nth device at time j, represents latent space data, represents the latent space data corresponding to the kth moment in the dth sensor in the nth device, represents the second layer projection vector, represents a nonlinear activation function, represents the second bias term, represents the first layer linear transformation weight, represents the second bias term, Represents the input data, which represents the sensor observation vector under each device n and sensor d. The superscript T represents the transpose operation, and the subscripts j and k are both time ordinal numbers. , m represents the latent space dimension.

[0034] Based on the potential feature data obtained by the encoder, it is mapped to the original space through the decoder. .in, Represents the decoder of the AE network.

[0035] In step S2 of the embodiment of the present invention, the loss function of the AE network is trained for: Where, represents the input data of the dth sensor in the nth device at time s, represents the reconstructed data of the d-th sensor in the n-th device at time s, represents the number of samples, Represents the number of electromechanical devices, and S represents the number of sensor types configured for each device.

[0036] In step S3 of the embodiment of the present invention, the OOD detector is used to evaluate the degree of deviation of the input sample relative to the ID data distribution, thereby realizing OOD data recognition. Specifically, the method of the OOD detector for OOD data recognition is: Calculate the reconstruction loss value of the input sample after passing through the parameter-frozen AE network, and compare the reconstruction loss value with the set discrimination threshold. When the reconstruction loss value exceeds the set discrimination threshold, the input sample is OOD data, otherwise it is ID data; The OOD data is abnormal data.

[0037] In this embodiment, a method based on statistical features is used to determine the discrimination threshold. The method for determining and setting the discrimination threshold is specifically as follows: For the training set containing only ID data, calculate the reconstruction loss value after passing through the AE network with frozen parameters; Count the mean of the reconstruction loss values ​​of all ID data and standard deviation , and then determine the discrimination threshold .

[0038] In this embodiment, the discrimination threshold determined by the above-mentioned "three sigma principle" method can accurately distinguish between data within the distribution and data outside the distribution, thereby achieving a higher detection capability while maintaining a low false alarm rate.

[0039] The "three sigma principle" is a method for identifying outliers. Its basic idea is that, under the assumption of a normal distribution, the vast majority (approximately 99.7%) of data will fall within the range of the mean plus or minus three standard deviations. Therefore, when training an AE network using only ID samples, its reconstruction error should be statistically concentrated within this range. Therefore, if the reconstruction error of a test sample is significantly higher than this range, it is very likely that its feature distribution differs significantly from the training data, and it can be reasonably determined to be an OOD sample.

[0040] In addition, this method has good versatility and interpretability in practical applications. On the one hand, since this method does not rely on any out-of-distribution data or label information, it is particularly suitable for unsupervised or semi-supervised scenarios; on the other hand, the threshold mechanism constructed through statistics also enhances the stability and robustness of the detection process, and can maintain consistent detection results in different data sets or application scenarios. It is worth noting that although the standard deviation principle theoretically assumes that the reconstruction error follows a normal distribution, in practical applications, even if the error distribution does not completely conform to the normal distribution, this method still has a good approximation effect, because it utilizes the quantitative relationship between the central tendency of the distribution and the degree of outliers, thereby effectively distinguishing normal samples from potential abnormal samples.

[0041] In summary, this embodiment adopts the The discrimination threshold setting method of "not only has a clear theoretical basis and is easy to implement, but also shows strong practical value under the AE network architecture. It is an effective strategy in unsupervised anomaly detection.

[0042] In step S4 of the embodiment of the present invention, the method for locating abnormal devices is specifically as follows: For input samples identified as abnormal, the reconstruction loss values ​​of all sensor channels on each device are accumulated item by item, and then the total reconstruction loss value in the device dimension is calculated. The device with the largest total reconstruction loss value is regarded as the abnormal device to be located.

[0043] Among them, the total reconstruction loss value for: In the formula, in the formula, represents the input data of the dth sensor in the nth device at time s, represents the reconstructed data of the dth sensor in the nth device at time s. This formula means that among all device dimensions, the device with the largest sum of reconstruction errors (i.e., the squared difference between the original value and the reconstructed value) is selected as the target device most likely to have an anomaly in the current sample.

[0044] Specifically, in this embodiment, according to the reconstruction error detection result, when the input sample at a certain moment When an anomaly is detected, it indicates that one or more electromechanical devices in the system may be operating abnormally. However, the detection results at this point only provide global information about the presence of an anomaly and cannot clearly pinpoint the specific device at the source of the fault. To further improve the system's diagnostic accuracy and localization capabilities, a detailed analysis of the reconstruction errors across each device dimension within the sample is required to identify the specific device most likely to have failed.

[0045] For each input sample that is judged to be abnormal , respectively on each device All sensor channels on The reconstruction errors are accumulated item by item, and the total reconstruction loss in the device dimension is calculated. The device with the largest total reconstruction loss value is regarded as the abnormal device to be located.

[0046] The core idea of ​​the above-mentioned anomaly localization method is that when a potential fault occurs in a piece of electromechanical equipment, its sensor characteristics often exhibit characteristics that are significantly different from the normal distribution in one or more modes, resulting in a significant increase in the reconstruction error in the corresponding dimension. By aggregating the reconstruction loss values ​​of all modes in the device dimension, the abnormal signals at the modal level can be effectively fused, thereby enhancing the robustness and reliability of anomaly localization.

[0047] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

[0048] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A method for identifying anomalies and predicting faults in electromechanical equipment based on multimodal OOD detection, characterized in that: The following steps are involved: S1, collects multimodal sensor time series data of the electromechanical equipment cluster and represents it as a three-dimensional data structure; The multimodal sensor time series data is ID data without any abnormality; S2. Use ID data to train the AE network and freeze the trained AE network parameters; S3. Based on the AE network with frozen parameters, an OOD detector is constructed; S4. Use the OOD detector to identify anomalies in the input samples of the three-dimensional data structure, and locate abnormal devices based on the reconstruction loss value of each modal sensor in the device dimension to achieve anomaly identification and fault prediction.

2. The method for identifying anomalies and predicting faults of electromechanical equipment based on multimodal OOD detection according to claim 1 is characterized in that: In step S1, the three-dimensional data structure of the sensor time series data is represented as (N, D, S), where N is the number of samples, indicating the number of observations collected at a set time or time period, D is the number of electromechanical devices, indicating the total number of independent devices in the electromechanical device cluster, and S indicates the number of sensor types configured for each device.

3. The method for identifying anomalies and predicting faults of electromechanical equipment based on multimodal OOD detection according to claim 1, characterized in that: In step S2, the AE network includes an encoder and a decoder that maintain a symmetric structure; The method for training the AE network is specifically as follows: The input data is mapped to a latent space by an encoder, and the latent feature data is mapped back to the original space by a decoder to obtain reconstructed data; wherein the latent feature data in the latent space is fused with multimodal feature information of the ID data; The reconstruction error between the input ID data and the reconstructed data is used as the loss function to train the AE network.

4. The method for identifying anomalies and predicting faults of electromechanical equipment based on multimodal OOD detection according to claim 3 is characterized in that: The process of mapping input data to latent space through the encoder is as follows: Perform linear transformation on the input data and convert it into latent space data; According to the attention function learned from the latent space data, a weight is assigned to each modality of the input data to obtain its corresponding assignable weight; The normalized attention weight is obtained by softmax operation on the assignable weight; Use attention weights to perform weighted summation on latent space data to obtain fused feature data that fuses multimodal feature information; The fused feature data is mapped to the latent space using a fully connected layer to obtain the latent feature data.

5. The method for identifying anomalies and predicting faults of electromechanical equipment based on multimodal OOD detection according to claim 4 is characterized in that: The potential feature data Expressed as: Where, represents the fusion feature data, represents the weights for mapping fused feature data to the latent space, represents the third bias term, represents the relu activation function, represents the normalized attention weight of the kth moment in the dth sensor in the nth device, represents the assignable weight of the dth sensor in the nth device, represents the allocable weight of the kth sensor in the nth device at time k, represents the assignable weight of the jth sensor in the nth device at time j, represents latent space data, represents the latent space data corresponding to the kth moment in the dth sensor in the nth device, represents the second layer projection vector, represents a nonlinear activation function, represents the second bias term, represents the first layer linear transformation weight, represents the second bias term, Represents the input data, which represents the sensor observation vector under each device n and sensor d. The superscript T represents the transpose operation, and the subscripts j and k are both time ordinal numbers. .

6. The method for identifying anomalies and predicting faults of electromechanical equipment based on multimodal OOD detection according to claim 3, characterized in that: In step S2, the loss function of the AE network is trained for: Where, represents the input data of the dth sensor in the nth device at time s, represents the reconstructed data of the d-th sensor in the n-th device at time s, represents the number of samples, Represents the number of electromechanical devices, and S represents the number of sensor types configured for each device.

7. The method for identifying anomalies and predicting faults of electromechanical equipment based on multimodal OOD detection according to claim 1, characterized in that: In step S3, the OOD detector is used to evaluate the degree of deviation of the input sample relative to the ID data distribution, thereby realizing OOD data recognition; The method for the OOD detector to perform OOD data recognition is: Calculate the reconstruction loss value of the input sample after passing through the parameter-frozen AE network, and compare the reconstruction loss value with the set discrimination threshold. When the reconstruction loss value exceeds the set discrimination threshold, the input sample is OOD data, otherwise it is ID data; The OOD data is abnormal data.

8. The method for identifying anomalies and predicting faults of electromechanical equipment based on multimodal OOD detection according to claim 7 is characterized in that: The method for determining the discrimination threshold is as follows: For the training set containing only ID data, calculate the reconstruction loss value after passing through the AE network with frozen parameters; Count the mean of the reconstruction loss values ​​of all ID data and standard deviation , and then determine the discrimination threshold .

9. The method for identifying anomalies and predicting faults of electromechanical equipment based on multimodal OOD detection according to claim 7, characterized in that: In step S4, the method for locating abnormal devices is specifically as follows: For input samples identified as abnormal, the reconstruction loss values ​​of all sensor channels on each device are accumulated item by item, and then the total reconstruction loss value in the device dimension is calculated. The device with the largest total reconstruction loss value is regarded as the abnormal device to be located.

10. The method for identifying anomalies and predicting faults of electromechanical equipment based on multimodal OOD detection according to claim 9, characterized in that: The total reconstruction loss value for: Where, represents the input data of the dth sensor in the nth device at time s, Represents the reconstructed data of the d-th sensor in the n-th device at time s.

Images