Engine anomaly detection method oriented to multi-source heterogeneous data and related products
By employing feature extraction and fusion methods for multi-source heterogeneous data and utilizing techniques such as dimension unification modules and attention networks, the problem of anomaly detection in multi-source heterogeneous data was solved, achieving high-precision engine anomaly detection and early warning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-17
- Publication Date
- 2026-03-24
AI Technical Summary
Existing technologies struggle to effectively integrate heterogeneous data from multiple sources, such as vibration, speed, and temperature, resulting in insufficient sensitivity and reliability in engine anomaly detection.
By employing a dimension unification module, attention network, softmax layer, reconstruction network, and outlier calculation module, we can achieve the fusion of multi-source heterogeneous data and outlier calculation by extracting features and mapping them to a latent space of a unified dimension, calculating adaptive importance weights, and reconstructing features.
It achieves high-precision and robust engine anomaly detection and early warning, improving the accuracy and reliability of anomaly detection.
Smart Images
Figure CN121720722A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of engine anomaly detection technology, and more specifically to engine anomaly detection methods and related products for multi-source heterogeneous data. Background Technology
[0002] Currently, engine anomaly detection mainly relies on the analysis and modeling of its operating status data. Common methods include monitoring techniques based on single data sources such as vibration signals, speed signals, or temperature signals. These methods typically design feature extraction and anomaly judgment rules for specific types of data, and can reflect local state changes of the engine to a certain extent.
[0003] However, due to the complexity and variable operating conditions of engine systems, information from a single data source often has limitations and cannot comprehensively and accurately characterize the overall health of the system. Furthermore, multi-source data (such as vibration, speed, and temperature) exhibit significant differences in data structure, dimensions, and feature distribution, representing typical heterogeneous multi-source data. Directly splicing or fusing these data can easily lead to information redundancy, noise interference, or dilution of important features, thus affecting the sensitivity and reliability of anomaly detection. Therefore, how to provide an engine anomaly detection method and related products for multi-source heterogeneous data, which can effectively fuse multi-source heterogeneous time-series information such as vibration, speed, and temperature, and achieve high-precision, robust anomaly detection and early warning, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0004] In view of the above problems, the present invention is proposed to provide an engine anomaly detection method and related products for multi-source heterogeneous data that overcomes or at least partially solves the above problems.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a method for detecting engine anomalies based on multi-source heterogeneous data includes the following steps: S1: Acquire multi-source heterogeneous data of the engine; wherein, the multi-source heterogeneous data includes vibration time-series signal, speed time-series signal and temperature time-series signal; S2: After feature extraction from the multi-source heterogeneous data, the data is input into the engine anomaly detection model to obtain the engine's anomaly value; wherein, the engine anomaly detection model includes a dimension unification module, an attention network, a softmax layer, a reconstruction network, and an anomaly value calculation module connected in sequence; the dimension unification module includes several parallel embedding networks; the anomaly value calculation module uses the similarity between the features before and after reconstruction by the reconstruction network to calculate the engine's anomaly value; S3: Determine whether the abnormal value exceeds a preset threshold. If it does, issue an engine abnormality warning.
[0006] Preferably, S2 specifically includes the following steps: S21: Perform feature extraction on the vibration time-series signal, the rotational speed time-series signal, and the temperature time-series signal to obtain a first feature vector, a second feature vector, and a third feature vector; S22: Using three independent embedding networks, the first feature vector, the second feature vector, and the third feature vector are mapped to a latent space of the same dimension to obtain the first latent feature, the second latent feature, and the third latent feature; S23: The first hidden feature, the second hidden feature, and the third hidden feature are processed sequentially through an attention network and a Softmax layer to obtain the adaptive importance weights of the vibration signal, the speed signal, and the temperature signal. S24: Obtain fused features based on the adaptive importance weights of the vibration signal, the adaptive importance weights of the rotation speed signal, the adaptive importance weights of the temperature signal, the first hidden feature, the second hidden feature, and the third hidden feature; S25: Input the fused features into the reconstruction network to obtain the reconstruction features; S26: Calculate outliers based on the reconstructed features and the fusion features.
[0007] Preferably, S21 specifically includes the following steps: Extract the time-domain features, frequency-domain features, and time-frequency features of the vibration time-series signal to obtain the first feature vector; Extract the statistical features of the rotational speed timing signal to obtain the second feature vector; The statistical features of the temperature time series signal are extracted to obtain the third feature vector.
[0008] Preferably, the time-domain features include root mean square and kurtosis; The frequency domain characteristics include spectral peaks and bandwidth energy; The time-frequency features include wavelet packet energy entropy; The statistical characteristics include mean and variance.
[0009] Preferably, the fusion feature is obtained based on the following formula: ; in, This represents the fusion feature; This represents the adaptive importance weight of the vibration signal; The adaptive importance weight represents the rotational speed signal; The adaptive importance weights of the temperature signal are represented. This represents the first hidden feature; This represents the second hidden feature; This represents the third hidden feature; This indicates element-wise multiplication; This represents the trainable ratio coefficient.
[0010] Preferably, the outliers are obtained based on the following formula: ; in, This indicates the outlier value; This represents the reconstructed feature; express The square of; Expressing the request The L2 norm.
[0011] Preferably, the reconstructed network is a stacked autoencoder.
[0012] Preferably, the embedded network is a fully connected layer.
[0013] In a second aspect, an engine anomaly detection system for multi-source heterogeneous data is provided to implement the engine anomaly detection method for multi-source heterogeneous data described in the first aspect, including a data acquisition module, a feature extraction module, and an engine anomaly detection model. The data acquisition module is used to acquire multi-source heterogeneous data of the engine; wherein, the multi-source heterogeneous data includes vibration time-series signals, speed time-series signals and temperature time-series signals; The feature extraction module is used to extract features from the multi-source heterogeneous data; The engine anomaly detection model is used to process the features extracted by the feature extraction module to obtain engine anomaly detection results.
[0014] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the engine anomaly detection method for multi-source heterogeneous data as described in the first aspect.
[0015] As can be seen from the above technical solution, compared with the prior art, the present invention discloses an engine anomaly detection method and related products for multi-source heterogeneous data, which can effectively integrate multi-source heterogeneous data such as vibration, speed, and temperature, and achieve high-precision and robust anomaly detection and early warning. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0017] Figure 1 This is a flowchart of an engine anomaly detection method for multi-source heterogeneous data provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] Firstly, such as Figure 1 As shown in the figure, this invention discloses a method for detecting engine anomalies based on multi-source heterogeneous data, including the following steps: S1: Acquire multi-source heterogeneous data of the engine; wherein, the multi-source heterogeneous data includes vibration time-series signal, speed time-series signal and temperature time-series signal; It is understandable that: Multi-source heterogeneous data is acquired through a sensor group and data acquisition system deployed on the engine; the sensor group includes vibration sensors, speed sensors and temperature sensors; the data acquisition system performs signal conditioning and synchronous analog-to-digital conversion on the analog signals output by each sensor to generate synchronous vibration timing signals, speed timing signals and temperature timing signals.
[0020] Signal conditioning includes: performing anti-aliasing low-pass filtering and amplification on the analog signal output by the vibration sensor; waveform shaping on the analog signal output by the speed sensor; and linearization compensation on the temperature signal output by the temperature sensor.
[0021] Synchronous analog-to-digital conversion refers to the use of a multi-channel synchronous acquisition card to sample the analog signals output by each sensor under a unified clock and trigger.
[0022] S2: After feature extraction from multi-source heterogeneous data, the data is input into the engine anomaly detection model to obtain the engine's anomaly value. The engine anomaly detection model includes a dimension unification module, an attention network, a softmax layer, a reconstruction network, and an anomaly value calculation module connected in sequence. The dimension unification module includes several parallel embedding networks. The anomaly value calculation module uses the similarity between the features before and after reconstruction by the reconstruction network to calculate the engine's anomaly value. In one embodiment, S2 specifically includes the following steps: S21: Perform feature extraction on the vibration time-series signal, rotational speed time-series signal, and temperature time-series signal to obtain the first feature vector, the second feature vector, and the third feature vector; In one embodiment, S21 specifically includes the following steps: Extract the time-domain features, frequency-domain features, and time-frequency features of the vibration time-series signal to obtain the first feature vector; Extract the statistical features of the rotational speed time-series signal to obtain the second feature vector; Statistical features of the temperature time series signal are extracted to obtain the third feature vector.
[0023] In one embodiment, the temporal features include root mean square and kurtosis; Frequency domain characteristics include spectral peaks and band energy; Time-frequency features include wavelet packet energy entropy; Statistical characteristics include mean and variance.
[0024] S22: Using three independent embedding networks, the first feature vector, the second feature vector, and the third feature vector are mapped to a latent space of the same dimension to obtain the first latent feature, the second latent feature, and the third latent feature; Specifically: the first feature vector is input into the first embedding network to obtain the first hidden feature; The second feature vector is input into the second embedding network to obtain the second hidden feature; The third feature vector is input into the third embedding network to obtain the third hidden feature.
[0025] Three parallel and independent embedding networks map the first, second, and third feature vectors to a latent space of a unified dimension.
[0026] In one embodiment, the embedded network is a fully connected layer.
[0027] S23: The first hidden feature, the second hidden feature, and the third hidden feature are processed sequentially through an attention network and a Softmax layer to obtain the adaptive importance weights of the vibration signal, the speed signal, and the temperature signal. In one embodiment, S23 specifically includes the following steps: 1) Concatenate the first, second, and third latent features along the feature dimension to obtain a combined feature matrix; 2) Input the combined feature matrix into the multi-head self-attention network and then fuse the output features of each head to obtain the enhanced feature matrix; Multi-head self-attention networks compute the correlations between different signal features using multiple parallel attention heads. Specifically, within each attention head, a trainable linear transformation projects the combined feature matrix into a query matrix, a key matrix, and a value matrix, respectively. Attention weights are then calculated using the scaled dot product attention formula, and the value matrices are weighted and fused to obtain the output of that attention head. The outputs of all attention heads are concatenated along the feature dimension and fused through a trainable linear projection layer to finally obtain an enhanced feature matrix containing the dependencies between signals. 3) The enhanced feature matrix is mapped through a fully connected layer (whose weight parameters are trainable) to convert the enhanced feature matrix into a three-dimensional vector. The three elements of this three-dimensional vector correspond to the original importance scores of the vibration, rotational speed and temperature signals, respectively.
[0028] 4) Input the original importance scores of vibration, rotation speed and temperature signals into the Softmax layer for normalization processing to obtain the final weight allocation with a sum of 1, namely the adaptive importance weight of vibration signal, adaptive importance weight of rotation speed signal and adaptive importance weight of temperature signal.
[0029] S24: Based on the adaptive importance weight of vibration signal, adaptive importance weight of rotation speed signal, adaptive importance weight of temperature signal, first hidden feature, second hidden feature, and third hidden feature, the fused feature is obtained; In one embodiment, the fusion features are obtained based on the following formula: ; in, Indicates fusion characteristics; Represents the adaptive importance weights of vibration signals; The adaptive importance weights represent the rotational speed signals; Represents the adaptive importance weights of the temperature signal; Indicates the first hidden feature; Indicates the second hidden feature; Indicates the third hidden feature; This indicates element-wise multiplication; This represents the trainable ratio coefficient.
[0030] It is understandable that the trainable scaling factor is updated by minimizing the reconstruction error loss. .
[0031] S25: Input the fused features into the reconstruction network to obtain the reconstructed features; In one embodiment, the reconstructed network is a stacked autoencoder.
[0032] S26: Calculate outliers based on reconstructed features and fused features.
[0033] In one embodiment, outliers are obtained based on the following formula: ; in, Indicates outlier; Represents reconstructed features; express The square of; Expressing the request The L2 norm.
[0034] It is understandable that the pre-trained engine anomaly detection model is obtained based on the following steps: Obtain the training set; where each sample in the training set is a set of synchronized vibration time-series signals, rotational speed time-series signals, and temperature time-series signals; After extracting features from each sample (in the same way as S2 feature extraction), the samples are input into the engine anomaly detection model for training. The reconstruction error loss is minimized to obtain the pre-trained engine anomaly detection model.
[0035] S3: Determine whether the abnormal value exceeds the preset threshold. If it does, issue an engine abnormality warning.
[0036] Based on the following steps, a preset threshold is obtained: Collect training samples of the engine under normal conditions; each training sample is a set of synchronized vibration timing signals, speed timing signals and temperature timing signals; Calculate the outliers for each training sample to form a historical outlier sequence; Sort the historical outlier sequence and select the value corresponding to the specified quantile (such as 95% or 99%) as the outlier threshold.
[0037] The above-mentioned method for determining the preset threshold can effectively control the false alarm rate and adapt to the operating characteristics of different engines.
[0038] In a second aspect, embodiments of the present invention also provide an engine anomaly detection system for multi-source heterogeneous data, used to implement the engine anomaly detection method for multi-source heterogeneous data in the first aspect, including a data acquisition module, a feature extraction module, and an engine anomaly detection model; The data acquisition module is used to acquire multi-source heterogeneous data of the engine; the multi-source heterogeneous data includes vibration time-series signals, speed time-series signals and temperature time-series signals; The feature extraction module is used to extract features from multi-source heterogeneous data; The engine anomaly detection model is used to process the features extracted by the feature extraction module to obtain engine anomaly detection results.
[0039] Thirdly, the present invention also provides an electronic device, such as... Figure 2 As shown, the electronic device may include a processor 201, a communications interface 202, a memory 203, and a communication bus 204. The processor 201, communications interface 202, and memory 203 communicate with each other via the communication bus 204. The processor 201 can call logical instructions from the memory 203 to execute an engine anomaly detection method for multi-source heterogeneous data.
[0040] Furthermore, the logical instructions in the aforementioned memory 203 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0041] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0042] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for detecting engine anomalies based on multi-source heterogeneous data, characterized in that, Includes the following steps: S1: Acquire multi-source heterogeneous data of the engine; wherein, the multi-source heterogeneous data includes vibration time-series signal, speed time-series signal and temperature time-series signal; S2: After feature extraction from the multi-source heterogeneous data, the data is input into the engine anomaly detection model to obtain the engine's anomaly value; wherein, the engine anomaly detection model includes a dimension unification module, an attention network, a softmax layer, a reconstruction network, and an anomaly value calculation module connected in sequence; the dimension unification module includes several parallel embedding networks; the anomaly value calculation module uses the similarity between the features before and after reconstruction by the reconstruction network to calculate the engine's anomaly value; S3: Determine whether the abnormal value exceeds a preset threshold. If it does, issue an engine abnormality warning.
2. The engine anomaly detection method for multi-source heterogeneous data according to claim 1, characterized in that, S2 specifically includes the following steps: S21: Perform feature extraction on the vibration time-series signal, the rotational speed time-series signal, and the temperature time-series signal to obtain a first feature vector, a second feature vector, and a third feature vector; S22: Using three independent embedding networks, the first feature vector, the second feature vector, and the third feature vector are mapped to a latent space of the same dimension to obtain the first latent feature, the second latent feature, and the third latent feature; S23: The first hidden feature, the second hidden feature, and the third hidden feature are processed sequentially through an attention network and a Softmax layer to obtain the adaptive importance weights of the vibration signal, the speed signal, and the temperature signal. S24: Obtain fused features based on the adaptive importance weights of the vibration signal, the adaptive importance weights of the rotation speed signal, the adaptive importance weights of the temperature signal, the first hidden feature, the second hidden feature, and the third hidden feature; S25: Input the fused features into the reconstruction network to obtain the reconstruction features; S26: Calculate the outlier based on the reconstructed features and the fusion features.
3. The engine anomaly detection method for multi-source heterogeneous data according to claim 2, characterized in that, S21 specifically includes the following steps: Extract the time-domain features, frequency-domain features, and time-frequency features of the vibration time-series signal to obtain the first feature vector; Extract the statistical features of the rotational speed timing signal to obtain the second feature vector; The statistical features of the temperature time series signal are extracted to obtain the third feature vector.
4. The engine anomaly detection method for multi-source heterogeneous data according to claim 3, characterized in that: The temporal features include root mean square and kurtosis; The frequency domain characteristics include spectral peaks and bandwidth energy; The time-frequency features include wavelet packet energy entropy; The statistical characteristics include mean and variance.
5. The engine anomaly detection method for multi-source heterogeneous data according to claim 3, characterized in that, The fusion feature is obtained based on the following formula: ； in, This represents the fusion feature; This represents the adaptive importance weight of the vibration signal; The adaptive importance weight represents the rotational speed signal; The adaptive importance weights of the temperature signal are represented. This represents the first hidden feature; This represents the second hidden feature; This represents the third hidden feature; This indicates element-wise multiplication; This represents the trainable ratio coefficient.
6. The engine anomaly detection method for multi-source heterogeneous data according to claim 5, characterized in that, The outliers are obtained based on the following formula: ； in, This indicates the outlier value; This represents the reconstructed feature; express The square of; Expressing the request The L2 norm.
7. The engine anomaly detection method for multi-source heterogeneous data according to claim 2, characterized in that, The reconstructed network is a stacked autoencoder.
8. The engine anomaly detection method for multi-source heterogeneous data according to claim 2, characterized in that, The embedded network is a fully connected layer.
9. An engine anomaly detection system for multi-source heterogeneous data, characterized in that, The engine anomaly detection method for multi-source heterogeneous data as described in any one of claims 1-8 includes a data acquisition module, a feature extraction module, and an engine anomaly detection model. The data acquisition module is used to acquire multi-source heterogeneous data of the engine; wherein, the multi-source heterogeneous data includes vibration time-series signals, speed time-series signals and temperature time-series signals; The feature extraction module is used to extract features from the multi-source heterogeneous data; The engine anomaly detection model is used to process the features extracted by the feature extraction module to obtain engine anomaly detection results.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the engine anomaly detection method for multi-source heterogeneous data as described in any one of claims 1 to 8.