Method and system for diagnosing production quality problem of scroll compressor of automobile air conditioner

Abstract

The invention relates to the technical field of abnormal monitoring of automobile scroll compressors, in particular to a method and a system for diagnosing production quality problems of an automobile air conditioner scroll compressor. Comprising the steps that under a preset typical working condition, multi-source physical signals of the scroll compressor of the automobile air conditioner are synchronously collected, high-sampling-frequency signals and low-sampling-frequency signals are preprocessed, and preprocessed signals are obtained; carrying out segmentation processing on the preprocessed signal by adopting a sliding window mechanism; based on the time domain feature and the frequency domain feature, obtaining a high sampling frequency feature and a low sampling frequency feature; inputting the high sampling frequency features and the low sampling frequency features into a convolutional neural network, and performing weighted fusion in combination with an attention mechanism to obtain fusion features; and performing decision-level weighted fusion on the fusion features to generate a comprehensive diagnosis result. According to the method, the limitation of a traditional single signal detection method is overcome, the fault detection sensitivity and diagnosis accuracy are remarkably improved, and the anomaly recognition performance under the multi-source signals is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of abnormal monitoring technology for automotive scroll compressors, and in particular to a method and system for diagnosing production quality problems of automotive air conditioning scroll compressors. Background Technology

[0002] In recent years, due to the increasingly urgent issues of energy shortage and environmental pollution, and under the policy guidance and advocacy of environmental protection departments, my country's new energy vehicle industry has developed rapidly, with its output increasing year by year. Against this backdrop, the scroll compressor, as a core component of the air conditioning system of new energy vehicles, has become increasingly important. When the core components of a scroll compressor (such as the scroll plate and bearings) malfunction, it can lead to leakage in the gas compression chamber or a reduction in compression efficiency, significantly affecting the intake and exhaust pressures and temperatures, resulting in decreased compressor performance, weakened cooling effect, and severe damage to its NVH performance. Traditional detection methods for such anomalies typically rely on single indicators such as vibration or noise signals, using traditional time-domain statistical feature analysis and frequency-domain fault feature diagnosis, and comparing the results with baseline data from a normal prototype to determine if a fault exists. However, compressor anomalies usually manifest as coordinated changes in multiple signal characteristics such as vibration, noise, temperature, and pressure, exhibiting multimodal and coupled characteristics. While relying solely on a single type of signal or a few features for diagnosis can detect some abnormalities, it has significant limitations: on the one hand, it is difficult to comprehensively cover all the characteristics of different fault modes, and on the other hand, it is impossible to accurately distinguish fault types (such as wear and leakage, which may produce similar vibration or noise characteristics), thus limiting the comprehensiveness and reliability of the diagnostic system.

[0003] Therefore, a diagnostic method and system for manufacturing quality problems of automotive air conditioning scroll compressors are proposed to solve the above-mentioned problems. Summary of the Invention

[0004] The present invention aims to provide a method and system for diagnosing production quality problems of automotive air conditioning scroll compressors, so as to overcome the limitations of existing automotive air conditioning scroll compressor production quality problem diagnosis technologies that rely on a single or a small number of new feature sources for diagnosis. The technical problem to be solved by the present invention is achieved through the following technical solution.

[0005] The first aspect of this invention proposes a method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors, comprising: simultaneously acquiring multi-source physical signals of the automotive air conditioning scroll compressor under preset typical operating conditions, wherein the multi-source physical signals include vibration and noise signals of high sampling frequency, and temperature and pressure signals of low sampling frequency; preprocessing the high sampling frequency signals and the low sampling frequency signals respectively to obtain preprocessed signals; segmenting the preprocessed signals using a sliding window mechanism, and extracting time-domain features and frequency-domain features within each sliding window segment; based on the time-domain features and the frequency-domain features, using a feature filtering mechanism to obtain high sampling frequency features and low sampling frequency features; inputting the high sampling frequency features into a first convolutional neural network and combining them with an attention mechanism for weighted fusion to obtain a first fused feature; inputting the low sampling frequency features into a second convolutional neural network and combining them with an attention mechanism for weighted fusion to obtain a second fused feature; and performing decision-level weighted fusion of the first fused feature and the second fused feature to generate a comprehensive diagnostic result.

[0006] The second aspect of this invention provides a diagnostic system for production quality problems of automotive air conditioning scroll compressors, which implements the diagnostic method for production quality problems of automotive air conditioning scroll compressors described in the first aspect of this invention. The system includes: a data acquisition module, which simultaneously acquires multi-source physical signals of the automotive air conditioning scroll compressor under preset typical operating conditions, wherein the multi-source physical signals include vibration and noise signals at high sampling frequencies, and temperature and pressure signals at low sampling frequencies; a preprocessing module, which preprocesses the high-sampling-frequency signals and the low-sampling-frequency signals respectively to obtain preprocessed signals; and a second processing module, which uses a sliding window mechanism to process the preprocessed signals. The signal is segmented and processed, and time-domain and frequency-domain features are extracted within each sliding window segment. A frequency feature determination module, based on the time-domain and frequency-domain features, uses a feature filtering mechanism to obtain high-sampling-frequency features and low-sampling-frequency features. A fusion module inputs the high-sampling-frequency features into a first convolutional neural network and combines them with an attention mechanism for weighted fusion to obtain a first fused feature. The low-sampling-frequency features are input into a second convolutional neural network and combined with an attention mechanism for weighted fusion to obtain a second fused feature. A weighting module performs decision-level weighted fusion of the first and second fused features to generate a comprehensive diagnostic result.

[0007] A third aspect of the present invention provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs; wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method for diagnosing production quality problems of automotive air conditioning scroll compressors as described in the first aspect of the present invention.

[0008] A fourth aspect of the present invention provides a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method for diagnosing production quality problems of automotive air conditioning scroll compressors as described in the first aspect of the present invention.

[0009] The embodiments of the present invention have the following advantages: Compared with existing technologies, this invention overcomes the limitations of traditional single-signal detection methods by simultaneously acquiring and deeply fusing multi-source heterogeneous signals such as vibration, noise, temperature, and pressure. This significantly improves the sensitivity and accuracy of fault detection, effectively enhancing anomaly identification performance under multi-source signals. Simultaneously, by employing a sliding window segmentation mechanism, the system can simultaneously extract fault-sensitive features from both the time and frequency domains, fully exploring the discriminative information of signals in different domains, enhancing the ability to characterize the fault evolution process, and achieving complementary and efficient extraction of time-frequency domain features. Furthermore, this invention introduces domain prior knowledge in the feature fusion and decision-making stages, obtaining feature weights with clear physical meaning through weighted training. This improves the interpretability and reliability of the diagnostic process, providing interpretable and physically consistent fusion decisions.

[0010] Furthermore, this invention does not require hardware modifications to existing compressor systems and can be implemented directly using common sensors and acquisition equipment, significantly reducing system complexity and deployment costs. It is suitable for real-time status monitoring and intelligent operation and maintenance of air conditioning systems in new energy vehicles, and has good engineering applicability and promotional value. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating an example of the method for diagnosing manufacturing quality problems in automotive air conditioning scroll compressors according to the present invention. Figure 2 This is a schematic diagram of the time-domain index segmentation calculation process of the high and low sampling frequency signals of the scroll compressor in the diagnostic method for production quality problems of the scroll compressor of the present invention. Figure 3 This is a flowchart of the segmented calculation of frequency domain index of high and low sampling frequency signals of the scroll compressor in the diagnostic method for production quality problems of automotive air conditioning scroll compressors of the present invention. Figure 4 This is a schematic diagram illustrating the principle of the dual-channel high and low sampling frequency signal fusion method in the diagnostic method for production quality problems of automotive air conditioning scroll compressors according to the present invention; Figure 5 This is a structural block diagram of the automotive air conditioning scroll compressor production quality problem diagnosis system of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention; Figure 7This is a schematic diagram of a computer-readable medium embodiment according to the present invention. Detailed Implementation

[0012] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0013] In view of the above problems, this invention proposes a method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors. This method first simultaneously collects multi-source physical data, such as high-sampling-frequency vibration and noise signals and low-sampling-frequency temperature and pressure signals, from the automotive air conditioning scroll compressor under preset typical operating conditions. Targeted preprocessing is performed on the raw signals at different sampling frequencies to eliminate noise and interference. Next, a sliding window mechanism is used to segment the preprocessed signals, and time-domain and frequency-domain features are extracted from each window segment. Based on this, a feature filtering mechanism is used to identify and acquire key features at high and low sampling frequencies. These high-sampling-frequency features are input into a first convolutional neural network and weighted and fused using an attention mechanism to obtain a first fused feature; simultaneously, the low-sampling-frequency features are input into a second convolutional neural network and weighted and fused using an attention mechanism to obtain a second fused feature. Decision-level weighted integration is performed on these two sets of fused features to generate a comprehensive diagnostic result to determine the abnormal state of the compressor. Compared with existing technologies, this invention overcomes the limitations of traditional single-signal detection methods, significantly improves the sensitivity and diagnostic accuracy of fault detection, and effectively enhances the performance of anomaly identification under multi-source signals. Meanwhile, this invention enhances the ability to characterize the fault evolution process by using a sliding window segmentation processing mechanism, and can achieve time-frequency domain feature complementarity and efficient extraction.

[0014] Furthermore, this invention does not require hardware modifications to existing compressor systems and can be implemented directly using common sensors and acquisition equipment, significantly reducing system complexity and deployment costs. It is suitable for real-time status monitoring and intelligent operation and maintenance of air conditioning systems in new energy vehicles, and has good engineering applicability and promotional value.

[0015] It should be noted that this invention has wide applications, suitable for the entire lifecycle management of scroll compressors for new energy vehicle air conditioning systems, with a particular focus on quality control and factory inspection during the manufacturing stage. It provides compressor production lines with online, real-time intelligent diagnostic capabilities, enabling accurate identification and early warning of early or complex production defects, thereby effectively improving product quality, reducing rework rates, and lowering warranty costs. Furthermore, the method of this invention can also be extended to the status monitoring, fault prediction, and intelligent operation and maintenance of air conditioning systems during the service life of new energy vehicles. Through continuous online diagnostics, it ensures the reliability and performance of vehicle air conditioning systems and provides data support and decision-making basis for after-sales service.

[0016] Example 1 The following reference Figures 1 to 4 The present invention will be described in detail below.

[0017] Figure 1 This is a flowchart illustrating an example of the method for diagnosing manufacturing quality problems in automotive air conditioning scroll compressors according to the present invention.

[0018] like Figure 1 As shown, in step S1, under preset typical operating conditions, multi-source physical signals of the automotive air conditioning scroll compressor are synchronously acquired. The multi-source physical signals include vibration signals and noise signals with high sampling frequency, as well as temperature signals and pressure signals with low sampling frequency.

[0019] In a specific application example, taking the scroll compressor of a new energy vehicle as an example, the specific application process of synchronously acquiring multi-source physical signals of the scroll compressor of the automotive air conditioning system is explained in detail.

[0020] On a dedicated compressor control panel, the operating status of the new energy scroll compressor is precisely controlled according to preset operating conditions. The compressor is installed in a semi-anechoic chamber to effectively isolate environmental noise interference. After the operating conditions stabilize, a high-precision data acquisition system is used to synchronously acquire multi-channel raw signals at a predetermined sampling frequency, obtaining multi-source heterogeneous signals of the new energy scroll compressor under operating conditions, including vibration, noise, temperature, and pressure, providing a data foundation for subsequent anomaly diagnosis and analysis.

[0021] The aforementioned multi-source heterogeneous signals include, but are not limited to, the following high-sampling-frequency signals and low-sampling signals, wherein: The high-sampling-frequency signals include vibration and noise signals. Vibration signals are acquired by vibration sensors. These sensors, conforming to the ISO 10816 series standards, are positioned at the connection between the compressor housing and the vehicle assembly to measure vibration signals in the X, Y, and Z directions. Noise signals are acquired by noise sensors. Following the GB / T 18505 standard for testing automotive air conditioning compressors, one noise sensor is placed at a distance of 30 cm from the surface in three directions: the top, front, and left side of the compressor. The sampling frequency is set to 12800 Hz.

[0022] The low sampling frequency signals include temperature and pressure signals. These signals are acquired by temperature and pressure sensors, which are directly contacted and positioned at the compressor's suction and discharge ports, respectively, to ensure accurate monitoring of the refrigerant's thermodynamic state. The sampling frequency is set to 3200 Hz.

[0023] Before formal operation, each test compressor was uniformly charged with 50g of compressor oil and continuously charged with refrigerant (R134a) until no air bubbles appeared in the sight glass of the test bench, to ensure that the refrigerant charge fully met the test specifications. For normal compressors, valid data was collected for 20 seconds after 15 minutes of stable operation; for abnormal compressors, 20 seconds of data were also collected after either exhibiting obvious abnormalities during operation or after 15 minutes of continuous operation (whichever occurred first).

[0024] It should be noted that abnormal compressors usually have difficulty maintaining the set steady-state operating parameters (such as pressure and temperature fluctuations exceeding the allowable range), or the time required to reach steady state is significantly prolonged. This phenomenon can be used as an auxiliary basis for abnormality identification.

[0025] Optionally, data from 50 compressors can be collected, including 20 normal compressors and 30 faulty compressors (10 each of moving and stationary disc faults, moving disc bearing faults, and combined faults) to construct a representative dataset.

[0026] It should be noted that the above is only an optional example and should not be construed as a limitation of the present invention.

[0027] Next, in step S2, the high sampling frequency signal and the low sampling frequency signal are preprocessed respectively to obtain the preprocessed signal.

[0028] In a specific example, taking into account the characteristics of multi-source heterogeneous signals, targeted preprocessing was performed on the high-frequency and low-frequency signals respectively: Specifically, for vibration and noise signals with high sampling frequencies (e.g., 12800 Hz), which typically contain a large amount of high-frequency random noise unrelated to compressor operation, traditional filtering methods easily lead to distortion in the fault-sensitive frequency band. This embodiment employs a Non-Local Means (NLMeans) denoising algorithm for processing. This algorithm borrows from image denoising principles, using weighted averaging based on the similarity of non-local regions in the signal. This effectively suppresses noise while preserving the signal's detailed features and fault-sensitive information to the maximum extent. The expression is as follows:

[0029] In the formula Indicates the position of the target pixel; This represents the pixel value at position j in the original noisy image.

[0030] On the other hand, for temperature and pressure signals with low sampling frequencies (e.g., 3200 Hz), which change slowly and are susceptible to measurement noise, this embodiment employs a moving average (MA) algorithm for smoothing. This method, by linearly averaging the signal within a sliding window, effectively suppresses high-frequency random fluctuations while preserving the low-frequency trend components of the signal, significantly improving the smoothness and reliability of the signal and providing stable input for subsequent feature extraction. Its expression is as follows:

[0031] In the formula, M is the window size of the moving average, and n is the index of the current data point; Let be the sampled value of the input signal at time nk.

[0032] Through the aforementioned targeted preprocessing strategies, the signal-to-noise ratio of vibration, noise, temperature, and pressure signals was significantly improved without losing as much effective information as possible, laying a high-quality data foundation for subsequent feature extraction and multi-source fusion.

[0033] The high and low sampling frequency signals were preprocessed separately to improve signal quality and suppress noise interference. For high sampling frequency vibration and noise signals, the NL-means (non-local mean) filtering algorithm was used for processing. For low sampling frequency temperature and pressure signals, the moving average algorithm was used for smoothing to suppress signal fluctuations and measurement noise and highlight their trend changes.

[0034] It should be noted that the above is only an optional example and should not be construed as a limitation of the present invention.

[0035] Next, in step S3, the preprocessed signal is segmented using a sliding window mechanism, and time-domain and frequency-domain features are extracted within each sliding window segment.

[0036] In a specific example, in order to effectively capture the features of the preprocessed non-stationary signal, a sliding window mechanism is used to segment the preprocessed multi-source signal and extract time-domain and frequency-domain features simultaneously within each sliding window segment.

[0037] Specifically, the preprocessed high and low sampling frequency signals are finely segmented using an overlapping sliding window algorithm to fully capture the dynamic characteristics and transient features of the signals.

[0038] The sliding window mechanism sets the window length to 1 / 20 of the total signal length and the movement step size to half of the window length.

[0039] Preferably, the window length is set to 1 second and the movement step is 0.5 seconds, i.e., the window overlap rate is 50%. This design can balance temporal resolution and feature continuity, avoiding transient feature smoothing due to excessively long windows, and preventing insufficient frequency resolution in frequency domain analysis due to excessively short windows.

[0040] Within each analysis window segment, the system performs the following feature extraction operations sequentially: Time-domain feature extraction: Comprehensive calculation of various time-domain parameters of the signal within the window to characterize its amplitude distribution, fluctuation characteristics, impulse characteristics, and energy distribution. These time-domain features include, but are not limited to: Time-domain statistical characteristics: mean, variance, standard deviation, root square magnitude, skewness, etc. Waveform metrics: kurtosis, waveform factor, impulse factor, etc.; Energy metrics: root mean square value, average power, energy entropy, etc.

[0041] These indicators can capture the dynamic changes of signals over different time periods, reveal the potential fault evolution process, and lay the foundation for subsequent frequency domain feature extraction and multi-source fusion.

[0042]

[0043] Frequency domain feature extraction: Perform a fast Fourier transform on each window segment of the signal to convert the time domain signal into a frequency domain representation, thereby obtaining spectral data that reflects the frequency structure and energy distribution of the signal.

[0044] Preferably, in order to reduce spectral leakage and improve frequency resolution, the Hanning window function is used to window the signal, and further optimization is performed using zero-filling technology.

[0045] A set of frequency domain feature indicators with clear physical meaning is extracted from the spectrum of each window to quantify the spectral characteristics of the signal. These indicators mainly include: Spectral center characteristics: centroid frequency, mean square frequency, frequency standard deviation, etc.

[0046]

[0047] By extracting the time-domain and frequency-domain features in a coordinated manner, the operating status of the scroll compressor can be fully perceived from multiple dimensions, and the discrimination information of the signal in different domains can be fully explored.

[0048] It should be noted that the above is only an optional example and should not be construed as a limitation of the present invention.

[0049] Next, in step S4, based on the time-domain features and the frequency-domain features, a feature filtering mechanism is used to obtain high sampling frequency features and low sampling frequency features.

[0050] Based on the time-domain and frequency-domain features obtained from S3, a multi-criteria feature selection mechanism is employed to obtain a subset of key features with high discriminative power and low redundancy, namely high-sampling-frequency features and low-sampling-frequency features. This selection process aims to optimize the feature indicators that are highly sensitive to and discriminative of abnormal states of scroll compressors, providing high-quality and refined data for the input of subsequent deep learning models.

[0051] In a specific example, a multi-stage screening strategy that comprehensively evaluates the statistical significance, information relevance, and redundancy of features is adopted. The process is as follows: First, based on information theory methods, the mutual information between all extracted high and low sampling frequency features and compressor abnormal state labels is calculated. The mutual information quantifies the strength of the nonlinear correlation between features and categories. To take into account the characteristics of signals in different frequency domains, this embodiment sets differentiated mutual information filtering thresholds: For high sampling frequency features (such as features originating from vibration signals or noise signals), a mutual information threshold of 0.25 is set, and only features with mutual information values ​​not lower than this threshold are retained.

[0052] For low sampling frequency features (such as features originating from temperature or pressure signals), a mutual information threshold of 0.15 is set, and only features with mutual information values ​​not lower than this threshold are retained.

[0053] This screening process effectively eliminates features with low correlation to abnormal states, achieving initial simplification of feature dimensions.

[0054] Significance test based on analysis of variance: For the features filtered by mutual information values, analysis of variance (ANOVA) was used to statistically test their distribution under different state categories (e.g., normal state, different fault types). The F-value and p-value of each feature were calculated (F-value and p-value are key indicators used in ANOVA in statistics), and features with statistically significant differences (p-value less than 0.05) between different operating conditions or states were selected. This stage aims to retain feature indicators that are highly sensitive to changes in compressor state and have strong discriminative ability.

[0055] Redundancy removal based on variance inflation factor: Furthermore, to eliminate redundancy within the feature subset and ensure independence between features, this embodiment employs an iterative elimination method based on variance inflation factor. This process calculates the variance inflation factor value for each feature in the feature set and continuously removes the feature with the highest current variance inflation factor value until the variance inflation factor values ​​of all remaining features are less than a preset threshold (preferably 10). This step ensures that the final feature subset not only possesses discriminative power but also maintains low internal correlation, thereby improving the generalization ability and robustness of subsequent models.

[0056] Domain knowledge and interpretability verification: Finally, by combining the failure mechanism of the scroll compressor with prior knowledge of the domain, the interpretability of the features obtained by the above statistical screening is verified to ensure that the selected features have clear physical meaning, thereby constructing a multi-source feature set with low redundancy, high sensitivity and strong generalization ability.

[0057] The formulas involved in the above process are as follows:

[0058]

[0059]

[0060] In the formula, It is the joint probability distribution of feature X and label Y. and It is a marginal probability distribution; It is the sum of squares between groups. is the sum of squares within groups, k is the number of categories, and N is the total number of samples.

[0061] Through the multi-criteria feature selection process described above, two independent feature matrices are finally obtained: high sampling frequency features (corresponding to vibration and noise signals) and low sampling frequency features (corresponding to temperature and pressure signals). Before inputting these selected features into the deep learning model, the resulting feature matrices are Z-score standardized to eliminate the influence of differences in feature dimensions and value ranges on model training.

[0062] The formula for calculating the Z-score using the normalization method is as follows:

[0063] In the formula, The original value, It is the mean of this feature across the samples. It is the standard deviation of the feature on the sample.

[0064] It should be noted that the above is only an optional example and should not be construed as a limitation of the present invention.

[0065] Furthermore, the accompanying drawings are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention and are not intended to be limiting. It is readily understood that the processes shown in the drawings do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0066] Next, in step S5, the high sampling frequency features are input into the first convolutional neural network and weighted and fused using an attention mechanism to obtain the first fused features; the low sampling frequency features are input into the second convolutional neural network and weighted and fused using an attention mechanism to obtain the second fused features. In a specific example, to fully utilize and deeply fuse high and low sampling frequency features, a dual-stream convolutional neural network (DS-CNN-CA) fusion model based on a cross-attention mechanism is adopted. This model aims to achieve deep interaction and complementary enhancement of high and low sampling frequency features, thereby more comprehensively and accurately diagnosing the compressor's operating status. The specific fusion process is as follows: After obtaining the filtered and standardized high-sampling-frequency feature matrix (derived from vibration signals and noise signals) and low-sampling-frequency feature matrix (derived from temperature signals and pressure signals), these two sets of feature matrices are respectively input into two independent convolutional neural network (CNN) branches in the DS-CNN-CA model.

[0067] The first convolutional neural network combines an attention mechanism (for high sampling frequency feature processing): a deep convolutional neural network (denoted as CNN-H) is constructed for high sampling frequency features. This CNN-H extracts local patterns and global contextual features of high-frequency signals layer by layer through multiple convolutional layers, activation functions, and pooling layers to obtain a high-level abstract representation.

[0068] The second convolutional neural network incorporates an attention mechanism (for low-sampling-frequency feature processing): a deep convolutional neural network (denoted as CNN-L) is constructed to handle low-sampling-frequency features. This CNN-L also extracts local patterns and global contextual features of low-frequency signals through multiple convolutional layers, activation functions, and pooling layers, thereby obtaining a high-level abstract representation.

[0069] Furthermore, after obtaining the abstract representations of the two branches, a cross-attention mechanism is introduced to achieve deep interaction of features: Low-frequency feature-guided high-frequency feature fusion: A set of feature vectors with clear physical meaning and high correlation to abnormal states is extracted and normalized using a fully connected layer and the Softmax function to generate an attention weight vector. This weight vector reflects the attention paid by low-frequency features to various parts of the high-frequency features. This weight vector is then element-wise weighted and fused with the high-sampling-frequency signal features output by CNN-H to obtain the first fused feature. This process, guided by the prior knowledge of the low-sampling-frequency signal features, enhances the fault-related sensitive components in the high-sampling-frequency signal features and suppresses irrelevant noise.

[0070] High-frequency features guide low-frequency feature fusion: Similarly, feature vectors are extracted, and an attention weight vector is generated. This weight vector is then element-wise weighted and fused with the low-sampling-frequency signal features output by CNN-L to obtain the second fused feature. This process allows the dynamic information of the high-sampling-frequency signal features to supplement and correct the low-sampling-frequency signal features.

[0071] Optionally, to train and evaluate the DS-CNN-CA fusion model, this embodiment uses a pre-defined dataset containing 50 scroll compressors, including data from 20 normal compressors and 30 faulty compressors. This dataset is randomly divided into a training set and a validation set in a 7:3 ratio: the training set contains data from 35 prototypes (14 normal compressors and 21 faulty compressors) for model parameter learning and optimization. The validation set contains data from the remaining 15 prototypes (6 normal compressors and 9 faulty compressors) to evaluate the model's generalization ability and diagnostic performance, and to guide the adjustment of model hyperparameters.

[0072] It should be noted that the above is only an optional example and should not be construed as a limitation of the present invention.

[0073] Furthermore, the accompanying drawings are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention and are not intended to be limiting. It is readily understood that the processes shown in the drawings do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0074] Next, in step S6, the first fusion feature and the second fusion feature are fused together at the decision level to generate a comprehensive diagnostic result.

[0075] Specifically, the decision-level weighted fusion uses a gradient descent-based optimization algorithm with anomaly identification accuracy as the objective function to solve for a set of optimal fusion weights. These optimal fusion weights are then applied to the first fusion feature and the second fusion feature for weighted integration to generate the comprehensive diagnostic result. This step aims to fully utilize the preliminary discrimination results from the high-frequency and low-frequency signal channels output by the dual-stream convolutional neural network (DS-CNN-CA) model in S5, and to achieve a more accurate, stable, and interpretable judgment of the abnormal state of the scroll compressor by optimizing the fusion weights.

[0076] In a specific example, after the DS-CNN-CA model is trained, each independent channel (high-sampling-frequency feature channel and low-sampling-frequency feature channel) outputs its corresponding fault category probability or anomaly score. To integrate these independent discrimination results, a decision-level weighted fusion strategy is adopted.

[0077] Through supervised training, an optimal set of fusion weights was obtained. The determination of these weights fully integrates prior knowledge of the failure mechanism of scroll compressors with data-driven characteristics. The specific optimization process is as follows: Based on the data from the 35 training prototypes defined in S5, an objective function is constructed to maximize the identification performance for specific fault types (e.g., wear failure of the moving and stationary discs, bearing failure of the moving disc, and combined failures of both). This objective function typically uses anomaly identification accuracy, F1 score, or cross-entropy loss as optimization metrics.

[0078] An optimization algorithm based on gradient descent (or combined with iterative optimization methods such as grid search) is used to iteratively solve the problem on the training set to find the optimal fusion weights that maximize the objective function. These weights represent the relative importance of the different channel discrimination results in the final diagnostic decision.

[0079] The optimal fusion weights obtained from the solution are applied to the outputs of the high-sampling-frequency channel and the low-sampling-frequency channel respectively for weighted integration. In this way, the optimal weighted combination based on the discrimination capability of each channel and its sensitivity to specific fault types is achieved, outputting a comprehensive anomaly score or fault category probability.

[0080] Ultimately, based on the maximum value of the comprehensive diagnostic probability or a preset threshold, the abnormal state of the compressor is accurately determined, thereby achieving more accurate, stable and interpretable compressor state diagnosis.

[0081] This method systematically completes the entire process from feature extraction and attention-weighted fusion to abnormal state identification by processing high- and low-sampling frequency signals from multiple heterogeneous sources. It achieves accurate and reliable diagnosis of abnormal states of scroll compressors in new energy vehicles, improving detection accuracy and robustness while possessing good engineering practicality and promotional value.

[0082] Example 2 The following are system embodiments of the present invention, which can be used to execute the method embodiments of the present invention. For details not disclosed in the system embodiments of the present invention, please refer to the method embodiments of the present invention.

[0083] Figure 5 This is a schematic diagram of an example of the new energy gear tooth surface quality detection and processing system of the present invention.

[0084] Reference Figure 5 The second aspect of this disclosure provides a diagnostic system for manufacturing quality problems of automotive air conditioning scroll compressors, employing the diagnostic method for manufacturing quality problems of automotive air conditioning scroll compressors described in the first aspect of this invention. The diagnostic system includes: The acquisition module synchronously acquires multi-source physical signals of the automotive air conditioning scroll compressor under preset typical operating conditions. The multi-source physical signals include vibration and noise signals with high sampling frequency, as well as temperature and pressure signals with low sampling frequency. The preprocessing module performs preprocessing on the high sampling frequency signal and the low sampling frequency signal respectively to obtain the preprocessed signal; The second processing module uses a sliding window mechanism to segment the preprocessed signal and extracts time-domain and frequency-domain features within each sliding window segment. The frequency feature determination module, based on time-domain and frequency-domain features, uses a feature filtering mechanism to obtain high-sampling-frequency features and low-sampling-frequency features; The fusion module inputs high-sampling-frequency features into a first convolutional neural network and performs weighted fusion using an attention mechanism to obtain a first fused feature; it also inputs low-sampling-frequency features into a second convolutional neural network and performs weighted fusion using an attention mechanism to obtain a second fused feature. The weighting module performs decision-level weighted fusion of the first and second fusion features to generate a comprehensive diagnostic result.

[0085] Figure 6 This is a schematic diagram of an embodiment of an electronic device according to the present invention.

[0086] like Figure 6 As shown, the electronic device is embodied in the form of a general-purpose computing device. There can be one or more processors working collaboratively. This invention also does not preclude distributed processing, meaning that processors can be distributed across different physical devices. The electronic device of this invention is not limited to a single entity, but can also be the sum of multiple physical devices.

[0087] The memory stores a computer-executable program, typically machine-readable code. The computer-readable program can be executed by the processor to enable the electronic device to perform the method of the present invention, or at least some steps of the method.

[0088] The memory includes volatile memory, such as random access memory (RAM) and / or cache memory, and may also be non-volatile memory, such as read-only memory (ROM).

[0089] Optionally, in this embodiment, the electronic device further includes an I / O interface for exchanging data with external devices. The I / O interface can represent one or more of several bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0090] It should be understood that Figure 6 The electronic device shown is merely one example of the present invention, and the electronic device of the present invention may also include elements or components not shown in the above examples. For example, some electronic devices also include display units such as displays, and some electronic devices also include human-computer interaction elements such as buttons and keyboards. Any electronic device capable of executing a computer-readable program in memory to implement the method of the present invention or at least some steps of the method can be considered as an electronic device covered by the present invention.

[0091] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software, or by combining software with necessary hardware. Therefore, as... Figure 7 As shown, the technical solution according to the embodiments of the present invention can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.) or on a network, and includes several commands to cause a computing device (such as a personal computer, server, or network device, etc.) to execute the above-described method according to the embodiments of the present invention.

[0092] The software product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0093] The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting programs for use by or in connection with a command execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0094] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0095] The aforementioned computer-readable medium carries one or more programs (e.g., computer executable programs) that, when executed by a device, cause the computer-readable medium to implement the methods of this disclosure.

[0096] Those skilled in the art will understand that the above modules can be distributed in the device as described in the embodiments, or they can be modified accordingly and placed in one or more devices that are unique to this embodiment. The modules in the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules.

[0097] Through the description of the above embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions of the embodiments of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, portable hard drive, etc.) or on a network, including several commands to cause a computing device (such as a personal computer, server, mobile terminal, or network device, etc.) to execute the methods according to the embodiments of the present invention.

[0098] Exemplary embodiments of the present invention have been specifically shown and described above. It should be understood that the present invention is not limited to the detailed structures, arrangements, or implementations described herein; rather, the present invention is intended to cover various modifications and equivalent arrangements contained within the spirit and scope of the appended claims.

Claims

1. A method for diagnosing manufacturing quality problems in automotive air conditioning scroll compressors, characterized in that, Includes the following steps: S1: Under preset typical operating conditions, synchronously collect multi-source physical signals of the automotive air conditioning scroll compressor. The multi-source physical signals include vibration and noise signals with high sampling frequency, as well as temperature and pressure signals with low sampling frequency. S2: Preprocess the high sampling frequency signal and the low sampling frequency signal respectively to obtain the preprocessed signal; S3: The preprocessed signal is segmented using a sliding window mechanism, and time-domain and frequency-domain features are extracted within each sliding window segment; S4: Based on the time-domain features and the frequency-domain features, a feature filtering mechanism is used to obtain high sampling frequency features and low sampling frequency features; S5: Input the high sampling frequency features into the first convolutional neural network and perform weighted fusion using an attention mechanism to obtain the first fused features; The low sampling frequency features are input into a second convolutional neural network and weighted and fused using an attention mechanism to obtain the second fused features. S6: Perform decision-level weighted fusion of the first fusion feature and the second fusion feature to generate a comprehensive diagnostic result.

2. The method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors according to claim 1, characterized in that, In step S2, the high-sampling-frequency signal and the low-sampling-frequency signal are preprocessed respectively, including: The high-sampling-frequency signal is denoised using the following expression: ； in, Indicates the position of the target pixel; This represents the pixel value at position j in the original noisy image.

3. The method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors according to claim 1, characterized in that, In step S2, the high-sampling-frequency signal and the low-sampling-frequency signal are preprocessed respectively, including: The low-sampling-frequency signal is denoised using the following expression: ； in, This represents the output signal of the filter at time n; M represents the window size of the moving average. This represents the sampled value of the input signal at time nk.

4. The method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors according to claim 1, characterized in that, In step S3, the sliding window mechanism sets the window length to 1 / 20 of the total signal length, and the movement step size to half of the window length.

5. The method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors according to claim 1, characterized in that, In step S3, extracting time-domain and frequency-domain features within each sliding window segment includes: The time-domain features include at least one of the following: mean, variance, root mean square magnitude, kurtosis, waveform factor, and root mean square value; The frequency domain features include at least one of the following: centroid frequency, mean square frequency, and frequency standard deviation.

6. The method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors according to claim 1, characterized in that, In step S4, the feature selection mechanism includes the following stages: The strength of the nonlinear correlation between each feature and the abnormal state is quantified by mutual information, while retaining features with high mutual information values. By calculating the frequency values ​​of each feature through analysis of variance, features that show significant differences between different states are selected. Analyze the redundancy between features and eliminate features with high information overlap and redundant contributions.

7. The method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors according to claim 1, characterized in that, In step S5, both the first convolutional neural network and the second convolutional neural network adopt a structure of alternating stacked convolutional layers and pooling layers to extract local patterns and global contextual features of the signal layer by layer.

8. The method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors according to claim 1, characterized in that, In step S6, the decision-level weighted fusion uses a gradient descent-based optimization algorithm with anomaly identification accuracy as the objective function to solve for a set of optimal fusion weights. The optimal fusion weights are then applied to the first fusion feature and the second fusion feature for weighted integration to generate the comprehensive diagnostic result.

9. The method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors according to claim 1, characterized in that, The abnormal state includes at least one of the following: wear failure of the moving and stationary discs, failure of the moving disc bearing, and a combined failure of the moving and stationary discs and bearings.

10. A diagnostic system for manufacturing quality problems of automotive air conditioning scroll compressors, characterized in that, The method for diagnosing manufacturing quality problems of automotive air conditioning scroll compressors according to any one of claims 1 to 9 includes: The acquisition module synchronously acquires multi-source physical signals of the automotive air conditioning scroll compressor under preset typical operating conditions. The multi-source physical signals include vibration and noise signals with high sampling frequency, as well as temperature and pressure signals with low sampling frequency. The preprocessing module preprocesses the high-sampling-frequency signal and the low-sampling-frequency signal respectively to obtain the preprocessed signal; The second processing module uses a sliding window mechanism to segment the preprocessed signal and extracts time-domain and frequency-domain features within each sliding window segment. The frequency feature determination module, based on the time domain features and the frequency domain features, uses a feature filtering mechanism to obtain high sampling frequency features and low sampling frequency features; The fusion module inputs the high-sampling-frequency features into a first convolutional neural network and performs weighted fusion using an attention mechanism to obtain a first fused feature; and inputs the low-sampling-frequency features into a second convolutional neural network and performs weighted fusion using an attention mechanism to obtain a second fused feature. The weighting module performs decision-level weighted fusion of the first fusion feature and the second fusion feature to generate a comprehensive diagnostic result.

Images