An engine fault detection system, method, apparatus, and storage medium

Abstract

The application provides an engine fault detection system, method, device and storage medium. At least three noise signals of an engine during quality inspection are obtained, preprocessed, and at least three target noise signals corresponding to a to-be-detected fault are obtained. Time-frequency conversion is performed to obtain a frequency energy spectrum feature corresponding to the to-be-detected fault. The frequency energy spectrum feature is sequentially subjected to normalization processing, feature enhancement processing and dimension reduction processing. The frequency energy spectrum feature after the dimension reduction processing is input into a fault recognition model to obtain a recognition result of whether the engine has the to-be-detected fault output by the fault recognition model. The fault recognition model is obtained based on deep residual network ResNet18 training. Compared with the prior art, the noise signals generated by the engine are analyzed and processed in detail to determine whether a specific type of fault exists in the engine, the influence of human error and subjective factors can be avoided, and the accuracy and efficiency of engine quality inspection are improved.

Description

Technical Field

[0001] This application relates to the field of engine technology, specifically to an engine fault detection system, method, device, and storage medium. Background Technology

[0002] Engine quality control is a crucial part of all engine manufacturing plants. Traditional quality inspection methods typically involve manual testing and inspection of various processes, including spark testing, fifth-gear testing, clutch testing, high-speed testing, idle speed testing, and oil spillage testing.

[0003] The aforementioned tests aim to assess the engine's performance and reliability under various operating conditions and identify any potential faults or problems. However, these methods are often limited by reliance on the subjective judgment of quality inspectors, inefficient testing processes, high costs of training professional quality inspectors, and human error. Therefore, improving the accuracy and efficiency of quality inspection and reducing human error is of paramount importance. Summary of the Invention

[0004] The purpose of this application is to provide an engine fault detection system, method, apparatus, and storage medium to improve the accuracy and efficiency of engine quality inspection.

[0005] The first aspect of this application provides an engine fault detection system, comprising: At least three noise sensors are set at different measuring points in the engine quality inspection station to collect noise signals of the engine during quality inspection. A multi-channel intelligent acquisition station, each acquisition channel of which corresponds to one of the noise sensors, is used to upload the noise signals acquired by at least three of the noise sensors to the central server; The central server is used to process at least three noise signals uploaded by the multi-channel intelligent acquisition station as follows: At least three noise signals are preprocessed to obtain at least three target noise signals corresponding to the fault to be tested; At least three target noise signals are converted from time to frequency to obtain the frequency energy spectrum characteristics corresponding to the fault under test. The frequency energy spectrum features are sequentially processed by normalization, feature enhancement, and dimensionality reduction. The frequency energy spectrum features after dimensionality reduction are input into the fault identification model to obtain the identification result of whether the engine has the fault to be tested. The fault identification model is trained based on the deep residual network ResNet18.

[0006] In one possible implementation, the central server performs feature enhancement processing on the frequency energy spectrum features in a manner that includes at least one of the following: time-domain shifting, random pruning, and spatial enhancement.

[0007] In one possible implementation, the central server performs dimensionality reduction processing on the frequency energy spectrum features, including: The frequency energy spectrum features were dimensionality reduced using principal component analysis.

[0008] In one possible implementation, the central server preprocesses at least three noise signals to obtain at least three target noise signals corresponding to the fault to be tested, including: For each noise signal, the test portion corresponding to the fault to be tested is extracted, and the test portion is organized into a preset data format to obtain the target noise signal.

[0009] In one possible implementation, the noise sensor has a frequency response range of 20Hz to 24kHz and a pickup parameter sensitivity of -45±4dB.

[0010] A second aspect of this application provides an engine fault detection method, comprising: Acquire at least three noise signals from the engine during quality inspection; At least three noise signals are preprocessed to obtain at least three target noise signals corresponding to the fault to be tested; At least three target noise signals are converted from time to frequency to obtain the frequency energy spectrum characteristics corresponding to the fault under test. The frequency energy spectrum features are sequentially processed by normalization, feature enhancement, and dimensionality reduction. The frequency energy spectrum features after dimensionality reduction are input into the fault identification model to obtain the identification result of whether the engine has the fault to be tested. The fault identification model is trained based on the deep residual network ResNet18.

[0011] In one possible implementation, the method of performing feature enhancement processing on the frequency energy spectrum features includes at least one of the following: time-domain shifting, random pruning, and spatial enhancement.

[0012] In one possible implementation, the dimensionality reduction processing of the frequency energy spectrum features includes: The frequency energy spectrum features were dimensionality reduced using principal component analysis.

[0013] A third aspect of this application provides an engine fault detection device, comprising: The acquisition module is used to acquire at least three noise signals of the engine during quality inspection; The preprocessing module is used to preprocess at least three noise signals to obtain at least three target noise signals corresponding to the fault to be tested; The time-frequency conversion module is used to perform time-frequency conversion on at least three target noise signals to obtain the frequency energy spectrum characteristics corresponding to the fault under test. The feature processing module is used to perform normalization, feature enhancement and dimensionality reduction processing on the frequency energy spectrum features in sequence. The identification module is used to input the frequency energy spectrum features after dimensionality reduction into the fault identification model to obtain the identification result of whether the engine has the fault to be tested, which is output by the fault identification model. The fault identification model is trained based on the deep residual network ResNet18.

[0014] A fourth aspect of this application provides a computer-readable storage medium having computer-readable instructions stored thereon, the computer-readable instructions being executable by a processor to implement the method as described in the second aspect.

[0015] This application discloses an engine fault detection system, method, apparatus, and storage medium. The system acquires at least three noise signals from the engine during quality inspection; preprocesses these noise signals to obtain at least three target noise signals corresponding to the fault to be detected; performs time-frequency conversion on the target noise signals to obtain the frequency energy spectrum features corresponding to the fault to be detected; sequentially performs normalization, feature enhancement, and dimensionality reduction processing on the frequency energy spectrum features; and inputs the dimensionality-reduced frequency energy spectrum features into a fault identification model to obtain the model's output indicating whether the engine has experienced the fault to be detected. The fault identification model is trained based on a deep residual network, ResNet18. Compared to existing technologies, this application, through detailed analysis and processing of the noise signals generated by the engine, determines whether a specific type of fault exists within the engine, avoiding human error and the influence of subjective factors, thereby improving the accuracy and efficiency of engine quality inspection. Attached Figure Description

[0016] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This paper shows a schematic diagram of the structure of an engine fault detection system provided in an embodiment of this application; Figure 2 The training accuracy and loss curve of the fault identification model of this application for rattling faults are shown; Figure 3 The training accuracy and loss curve of the fault identification model of this application for the time-lapse fault are shown; Figure 4 A flowchart of an engine fault detection method provided in an embodiment of this application is shown. Implementation

[0017] Exemplary embodiments of this application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.

[0018] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains.

[0019] Furthermore, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order. Additionally, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to those processes, methods, products, or apparatuses.

[0020] This application provides an engine fault detection system, method, device, and storage medium, which can improve the accuracy and efficiency of engine quality inspection. The following description is in conjunction with the accompanying drawings.

[0021] Please refer to Figure 1 It shows a schematic diagram of the structure of an engine fault detection system provided by some embodiments of this application, such as Figure 1 As shown, the engine fault detection system includes at least three noise sensors 10, a multi-channel intelligent acquisition station 20, and a central server 30. The multi-channel intelligent acquisition station 20 has multiple acquisition channels, and each acquisition channel is connected to a noise sensor 10.

[0022] The noise sensor is used to collect environmental noise signals and can transmit the noise signals to a multi-channel intelligent acquisition station in real time. The pickup parameters are: frequency response range 20Hz~24kHz, sensitivity: -45±4dB.

[0023] The multi-channel intelligent data acquisition station can be used to collect noise and vibration data, has edge computing capabilities, and transmits data to the central server in real time. It has the following hardware functions and performance indicators: a. Built-in acoustic and vibration analysis edge algorithm, supporting synchronous acquisition of multi-dimensional data such as IEPE (piezoelectric integrated circuit), voltage, current, audio, temperature, displacement, pressure, and switch quantity; b. It uses an Intel i5 processor and has edge computing capabilities; c. Memory capacity is 8G, and storage space is 512G; d. Supports multiple data communication interfaces, including 10 / 100 / 1000M communication interfaces and ModBus communication interfaces.

[0024] In this application, at least three noise sensors 10 are installed at different measuring points in the engine quality inspection station. These sensors can collect the noise signals emitted by the engine during quality inspection and send them to the multi-channel intelligent acquisition station 20. In other words, at least three noise sensors are configured at the engine quality inspection station to collect the noise emitted by the engine during quality inspection in real time. The different measuring points in the engine quality inspection station can be set according to the actual situation, such as being evenly arranged around the engine.

[0025] Each acquisition channel of the multi-channel intelligent acquisition station 20 corresponds to one of the noise sensors 10. It receives noise signals collected by at least three of the noise sensors and then uploads the noise signals collected by the at least three noise sensors to the central server 30. The central server 30 processes and analyzes the at least three noise signals uploaded by the multi-channel intelligent acquisition station 20 to determine whether there is a specific type of fault inside the engine, such as upward rattling, downward rattling, timing whistling, etc.

[0026] Specifically, the central server 30 processes at least three noise signals uploaded by the multi-channel intelligent acquisition station as follows: 1. Preprocess at least three of the noise signals to obtain at least three target noise signals corresponding to the fault to be tested.

[0027] Specifically, for each noise signal, the test portion corresponding to the fault to be tested is extracted, and the test portion is organized into a preset data format to obtain the target noise signal.

[0028] For example, the fault under test is a rattling fault, which includes upper and lower rattling. Upper rattling is caused by loose intake and exhaust valve nuts, leading to impact between the valve stem and the nut contact point. This fault can occur in the intake, exhaust, or dual valve systems. The difference between different rattling types lies in the location of the rattling and the number of times it occurs per cycle, which is related to the corresponding valve opening / closing. Lower rattling, on the other hand, has a more complex and varied cause, possibly due to missing teeth, wear, misalignment, or foreign objects falling into the engine's internal structure, affecting multiple related gears in the lower cylinder section. This type of fault occurs in the lower engine cavity and has a complex layout; the sound becomes reverberant after penetrating the engine sidewalls.

[0029] Quality inspection of rattling faults is mainly carried out through engine idling speed testing. Therefore, the idling test part corresponding to rattling faults in the noise signal can be extracted. Then, the raw data of the idling test part is organized into a data format with a 5-second sampling time and a 48,000Hz sampling rate. This step ensures the uniformity and consistency of the data, which is helpful for subsequent analysis and processing.

[0030] For example, timing squeal is a common engine malfunction, characterized by a high-frequency squealing sound during engine operation. Its formation is usually related to the engine's timing system. During engine operation, if the timing gears are too tight, too loose, or worn, it can cause problems with the timing system, resulting in a high-frequency squealing sound. The frequency of this squealing sound is usually related to the engine speed; the higher the speed, the higher the frequency. During an oil release test, the frequency increases when the throttle is briefly increased, and the timing squealing characteristic is obvious at the end of the oil release. The squealing frequency decreases from a high frequency after the oil release ends. Oil release is identified based on energy changes during oil release, and the left boundary of the squealing characteristic is determined when the oil release peak occurs.

[0031] The quality inspection of timing squeal fault is mainly carried out through the engine oil dispersal test. Therefore, the oil dispersal test part corresponding to the timing squeal fault in the noise signal can be extracted, and then the raw data of the oil dispersal test part can be organized into a preset data format.

[0032] 2. Perform time-frequency conversion on at least three of the target noise signals to obtain the frequency energy spectrum characteristics corresponding to the fault under test.

[0033] For example, performing a short-time Fourier transform (STFT) on the three target noise signals converts the time-domain waveforms into the frequency domain, extracting the frequency energy spectrum features (3*257*188*5) from the audio. The frequency characteristics of noise signals often contain rich information; for example, noise at a specific frequency may represent a specific type of fault.

[0034] 3. The frequency energy spectrum features are then subjected to normalization, feature enhancement, and dimensionality reduction processing in sequence.

[0035] Specifically, the frequency energy spectrum features can first be normalized. Normalization is a data preprocessing technique that eliminates the influence of dimensions on the data, allowing different features to be compared on the same scale. The normalized frequency energy spectrum features are then subjected to feature enhancement processing, including time-domain shifting, random pruning, and spatial enhancement. These operations aim to improve the model's generalization ability, enabling it to maintain good recognition performance even when dealing with slightly varied or noisy data.

[0036] Feature observation revealed that the shaking characteristics mainly occurred in the specified frequency band. Considering the sample size issue, Principal Component Analysis (PCA) was used to extract the main features of the data while removing noise and redundant information. Reducing the dimensionality of the aforementioned 257-length frequency domain features significantly reduced the complexity of data processing and model training, while also mitigating the risk of overfitting.

[0037] This complete data processing and feature extraction process can effectively extract useful features from raw audio data for identifying engine rattling problems.

[0038] The above process can be understood as a feature matrix analysis method. This method analyzes the frequency energy spectrum to determine the frequency spectrum of the fault, and a customized feature matrix can be obtained through a custom feature matrix processor. This feature matrix retains most of the time-domain features and a small number of frequency-domain features. By comparing the full-band spectral energy spectrum, it reduces the feature dimension, improves the convergence speed, and accelerates the network speed while retaining important features. Simultaneously, it adds spectral filtering to filter out unnecessary interference from low and mid-to-high frequencies. Compared to a pure time-domain one-dimensional feature, the custom feature matrix defined in this application still retains a small portion of the frequency-domain features, resulting in better performance than the pure time-domain one-dimensional feature in multiple experiments.

[0039] 4. Input the frequency energy spectrum features after dimensionality reduction into the fault identification model to obtain the identification result of whether the engine has the fault to be tested, which is output by the fault identification model. The fault identification model is trained based on the deep residual network ResNet18.

[0040] The training dataset can be obtained through the above data processing and feature extraction process. The fault identification model can then be obtained by training the deep residual network ResNet18 based on the training dataset.

[0041] For example, for rattling faults, after the above feature extraction and dimensionality reduction are completed, these features are fed into an improved ResNet network for rattling fault identification. The main advantages of the ResNet network lie in its deep architecture and the introduction of residual modules, which make it significantly superior in processing complex and high-dimensional data.

[0042] This application uses the deep residual network ResNet18 as the base model for fault identification. ResNet18 is a lightweight version of the ResNet series, containing 5 convolutional blocks (including 1 primary convolutional block and 4 deep residual blocks) and 1 fully connected layer, for a total of 18 layers. This network structure achieves a good balance between computational efficiency and model performance.

[0043] The network input is a preprocessed and extracted feature matrix of rattling sounds. First, a convolutional layer performs preliminary feature extraction. Then, these features are fed into four residual modules for layer-by-layer feature extraction and optimization. In the residual modules, a shortcut connection is used, allowing input features to propagate directly to subsequent layers, ensuring information integrity and continuity. Finally, a fully connected layer reduces the network's output dimensionality to the desired number of categories, outputting the probability of each category to achieve the identification of rattling-related faults.

[0044] The cross-entropy loss function is used during network training, a commonly used loss function for classification problems. The optimizer chosen is Adam, which demonstrates good performance in deep learning tasks. During training, an early stopping strategy is employed to prevent overfitting. Training stops when the loss on the validation set no longer decreases for several consecutive epochs, and the model at this point is saved as the final model. An epoch refers to the number of times the dataset is completely traversed during training.

[0045] Through this network design and training strategy, the fault identification model can effectively identify various engine faults and achieve high accuracy on the test set. This model can be further optimized by introducing more data and features to improve its generalization ability and stability.

[0046] For rattling-type faults, when using the idling stage energy spectrum + convolutional neural network method, the false alarm rate (falsely reported as faulty for qualified products) is 9.64%, the false negative rate (falsely reported as qualified for faulty products) is 5.51%, and the overall accuracy is 84.85%. This application uses a new principal component analysis algorithm to reduce redundant feature components, and the use of ResNet18 neural network can effectively alleviate the gradient vanishing and gradient exploding problems, thereby improving the model performance. In actual field detection results, this application achieved: a false alarm rate reduced to 5.21%, a false negative rate reduced to 1.23%, and an overall recognition rate increased to 93.56%. Figure 2 The figure shows the training accuracy and loss curve of the fault identification model of this application for rattling faults.

[0047] For timing howling faults, the identification features are relatively obvious, and the feature discrimination effect is quite good. Under ResNet network training, a test set identification accuracy of 96.5% can be achieved. In actual detection, the false alarm rate is 2.79%, the false negative rate is 0.90%, and the actual detection accuracy is 96.31%. Actual detection results demonstrate that the feature extraction part and the ResNet neural network part of timing howling have good identification effects on timing howling faults. Figure 3 The figure shows the training accuracy and loss curve of the fault identification model of this application for the timing howling fault.

[0048] Compared with existing technologies, the engine fault detection system of this application acquires at least three noise signals from the engine during quality inspection; preprocesses the at least three noise signals to obtain at least three target noise signals corresponding to the fault to be tested; performs time-frequency conversion on the at least three target noise signals to obtain the frequency energy spectrum features corresponding to the fault to be tested; sequentially performs normalization, feature enhancement, and dimensionality reduction processing on the frequency energy spectrum features; inputs the dimensionality-reduced frequency energy spectrum features into a fault identification model to obtain the identification result of whether the engine has the fault to be tested, which is trained based on a deep residual network ResNet18. Compared with existing technologies, this application, by performing detailed analysis and processing of the noise signals generated by the engine to determine whether a specific type of fault exists inside the engine, can avoid the influence of human error and subjective factors, thereby improving the accuracy and efficiency of engine quality inspection. This application also aims to provide engine manufacturing plants with a more reliable and efficient quality inspection method, thereby improving product quality, reducing failure rates, and enhancing consumer confidence in engine products.

[0049] Another embodiment of this application also provides an engine fault detection method, such as... Figure 4The diagram shown is a flowchart of the engine fault detection method provided in this application. The execution entity of this method can be the aforementioned central server, or other devices, such as... Figure 4 As shown, the engine fault detection method may include the following steps: S101. Obtain at least three noise signals from the engine during quality inspection; S102. Preprocess at least three noise signals to obtain at least three target noise signals corresponding to the fault to be tested; S103. Perform time-frequency conversion on at least three target noise signals to obtain the frequency energy spectrum characteristics corresponding to the fault under test; S104. The frequency energy spectrum features are sequentially processed by normalization, feature enhancement, and dimensionality reduction. S105. Input the frequency energy spectrum features after dimensionality reduction into the fault identification model to obtain the identification result of whether the engine has the fault to be tested, which is output by the fault identification model. The fault identification model is trained based on the deep residual network ResNet18.

[0050] In one possible implementation, the method of performing feature enhancement processing on the frequency energy spectrum features in step S104 includes at least one of the following: time-domain shifting, random cropping, and spatial enhancement.

[0051] In one possible implementation, step S104 involves dimensionality reduction of the frequency energy spectrum features, including: The frequency energy spectrum features were dimensionality reduced using principal component analysis.

[0052] In one possible implementation, step S102 involves preprocessing at least three noise signals to obtain at least three target noise signals corresponding to the fault to be tested, including: For each noise signal, the test portion corresponding to the fault to be tested is extracted, and the test portion is organized into a preset data format to obtain the target noise signal.

[0053] The engine fault detection method provided in this application is based on the same inventive concept and has the same beneficial effects as the engine fault detection system provided in the foregoing embodiments of this application.

[0054] Another embodiment of this application also provides an engine fault detection device, including: The acquisition module is used to acquire at least three noise signals of the engine during quality inspection; The preprocessing module is used to preprocess at least three noise signals to obtain at least three target noise signals corresponding to the fault to be tested; The time-frequency conversion module is used to perform time-frequency conversion on at least three target noise signals to obtain the frequency energy spectrum characteristics corresponding to the fault under test. The feature processing module is used to perform normalization, feature enhancement and dimensionality reduction processing on the frequency energy spectrum features in sequence. The identification module is used to input the frequency energy spectrum features after dimensionality reduction into the fault identification model to obtain the identification result of whether the engine has the fault to be tested, which is output by the fault identification model. The fault identification model is trained based on the deep residual network ResNet18.

[0055] In one possible implementation, the feature processing module performs feature enhancement processing on the frequency energy spectrum features in at least one of the following ways: time-domain shifting, random cropping, and spatial enhancement.

[0056] In one possible implementation, the feature processing module is specifically used to: perform dimensionality reduction processing on the frequency energy spectrum features using a principal component analysis algorithm.

[0057] In one possible implementation, the preprocessing module is specifically used to: for each noise signal, extract the test portion corresponding to the fault to be tested, and organize the test portion into a preset data format to obtain the target noise signal.

[0058] The engine fault detection device provided in this application embodiment is based on the same inventive concept as the engine fault detection system provided in the foregoing embodiments of this application and has the same beneficial effects.

[0059] This application also provides a computer-readable storage medium corresponding to the engine fault detection method provided in the foregoing embodiments, which stores a computer program (i.e., a program product) thereon. When the computer program is run by a processor, it executes the engine fault detection method provided in any of the foregoing embodiments.

[0060] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0061] The computer-readable storage medium provided in the above embodiments of this application and the engine fault detection method provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0062] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0063] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0064] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0065] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0066] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0067] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 described in the various embodiments of this application. 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.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and they should all be covered within the scope of the claims and specification of this application.

Claims

1. An engine fault detection system, characterized in that, include: At least three noise sensors are set at different measuring points in the engine quality inspection station to collect noise signals of the engine during quality inspection. A multi-channel intelligent acquisition station, each acquisition channel of which corresponds to one of the noise sensors, is used to upload the noise signals acquired by at least three of the noise sensors to the central server; The central server is used to process at least three noise signals uploaded by the multi-channel intelligent acquisition station as follows: At least three noise signals are preprocessed to obtain at least three target noise signals corresponding to the fault to be tested; At least three target noise signals are converted from time to frequency to obtain the frequency energy spectrum characteristics corresponding to the fault under test. The frequency energy spectrum features are sequentially processed by normalization, feature enhancement, and dimensionality reduction. The frequency energy spectrum features after dimensionality reduction are input into the fault identification model to obtain the identification result of whether the engine has the fault to be tested. The fault identification model is trained based on the deep residual network ResNet18. The central server performs feature enhancement processing on the frequency energy spectrum features in at least one of the following ways: time-domain shifting, random pruning, and spatial enhancement; the central server performs dimensionality reduction processing on the frequency energy spectrum features, including: using principal component analysis algorithm to reduce the dimensionality of the frequency energy spectrum features.

2. The engine fault detection system according to claim 1, characterized in that, The central server preprocesses at least three noise signals to obtain at least three target noise signals corresponding to the fault to be tested, including: For each noise signal, the test portion corresponding to the fault to be tested is extracted, and the test portion is organized into a preset data format to obtain the target noise signal.

3. The engine fault detection system according to claim 1, characterized in that, The noise sensor has a frequency response range of 20Hz to 24kHz and a pickup parameter sensitivity of -45±4dB.

4. An engine fault detection method, characterized in that, include: Acquire at least three noise signals from the engine during quality inspection; At least three noise signals are preprocessed to obtain at least three target noise signals corresponding to the fault to be tested; At least three target noise signals are converted from time to frequency to obtain the frequency energy spectrum characteristics corresponding to the fault under test. The frequency energy spectrum features are sequentially processed by normalization, feature enhancement, and dimensionality reduction. The frequency energy spectrum features after dimensionality reduction are input into the fault identification model to obtain the identification result of whether the engine has the fault to be tested. The fault identification model is trained based on the deep residual network ResNet18. The method for enhancing the frequency energy spectrum features includes at least one of the following: time-domain shifting, random pruning, and spatial enhancement; the method for reducing the dimensionality of the frequency energy spectrum features includes: using principal component analysis algorithm to reduce the dimensionality of the frequency energy spectrum features.

5. The engine fault detection method according to claim 4, characterized in that, The preprocessing of at least three noise signals to obtain at least three target noise signals corresponding to the fault to be tested includes: For each noise signal, the test portion corresponding to the fault to be tested is extracted, and the test portion is organized into a preset data format to obtain the target noise signal.

6. An engine fault detection device, characterized in that, include: The acquisition module is used to acquire at least three noise signals of the engine during quality inspection; The preprocessing module is used to preprocess at least three noise signals to obtain at least three target noise signals corresponding to the fault to be tested; The time-frequency conversion module is used to perform time-frequency conversion on at least three target noise signals to obtain the frequency energy spectrum characteristics corresponding to the fault under test. The feature processing module is used to perform normalization, feature enhancement and dimensionality reduction processing on the frequency energy spectrum features in sequence. The identification module is used to input the frequency energy spectrum features after dimensionality reduction into the fault identification model to obtain the identification result of whether the engine has the fault to be tested, which is output by the fault identification model. The fault identification model is trained based on the deep residual network ResNet18. The feature processing module performs feature enhancement processing on the frequency energy spectrum features in at least one of the following ways: time-domain shifting, random pruning, and spatial enhancement; the feature processing module performs dimensionality reduction processing on the frequency energy spectrum features, including: using principal component analysis algorithm to perform dimensionality reduction processing on the frequency energy spectrum features.

7. The engine fault detection device according to claim 6, characterized in that, The preprocessing module is used to: for each noise signal, extract the test portion corresponding to the fault to be tested, and organize the test portion into a preset data format to obtain the target noise signal.

8. A computer-readable storage medium, characterized in that, It stores computer-readable instructions that can be executed by a processor to implement the method as described in claim 4 or 5.

Images