Door closing sound anomaly diagnosis method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application provide a vehicle door closing sound abnormality diagnosis method and device, electronic equipment and storage medium. Among them, the method comprises: acquiring the original audio of the closing sound of the vehicle, and preprocessing the original audio of the closing sound to generate the target audio of the closing sound; the acoustic characteristics of the target audio of the closing sound are extracted to obtain the acoustic characteristics of the target audio of the closing sound, and a two-dimensional feature map is generated based on the acoustic characteristics; the two-dimensional feature map is inferred based on the closing sound abnormality diagnosis model to generate inference probability data and grade quantization data. Therefore, the automation and standardization of the whole process of the diagnosis of the closing sound of the vehicle door are realized, the subjective error and labor cost caused by manual listening and identification are effectively reduced, and the diagnosis efficiency and consistency are significantly improved. At the same time, through the comprehensive extraction and transformation of the acoustic characteristics, the recognition ability of the subsequent diagnosis link to various abnormalities, especially slight abnormal sound, is enhanced, and finally objective and quantitative judgment basis is provided for the production line.

Description

Technical Field

[0001] This application relates to the field of automotive manufacturing quality inspection technology, and in particular to a method, device, electronic device and storage medium for diagnosing abnormal door closing sounds. Background Technology

[0002] In the field of automotive manufacturing quality inspection, diagnosing abnormal door closing sounds is a core aspect of controlling the assembly precision and component quality of the entire vehicle. Abnormal door closing sounds often directly reflect assembly defects or component malfunctions such as hinge jamming, poor sealing strip assembly, abnormal door lock engagement clearance, sheet metal resonance, and loose interior parts. Timely and accurate diagnosis of abnormal door closing sounds can effectively reduce the after-sales failure rate of vehicles and improve product quality control. Therefore, an efficient and accurate door closing sound anomaly diagnosis solution is crucial for the high-precision, batch testing requirements of automotive production lines.

[0003] Currently, the industry primarily relies on two technical solutions for diagnosing abnormal door closing sounds: manual subjective diagnosis and traditional machine diagnosis. Manual subjective diagnosis suffers from high subjectivity, poor consistency of diagnostic results, low efficiency, and high costs. It struggles to generate standardized and traceable fault diagnosis conclusions, and the time-consuming nature of single-sample diagnosis makes it unsuitable for the batch testing needs of production lines. Traditional machine diagnosis, on the other hand, suffers from limitations such as limited feature sets, missing key information leading to low diagnostic accuracy, poor model generalization ability, weak scenario adaptability, and a single diagnostic dimension, failing to achieve fault tracing. Summary of the Invention

[0004] This application aims to at least partially address one of the technical problems in the related art.

[0005] Therefore, the first objective of this application is to propose a diagnostic method for abnormal door closing sounds, which automates and standardizes the entire diagnostic process, effectively reducing subjective errors and labor costs associated with manual listening, and significantly improving diagnostic efficiency and consistency. Simultaneously, through comprehensive extraction and transformation of acoustic features, it enhances the ability of subsequent diagnostic stages to identify various abnormalities, especially subtle noises, ultimately providing the production line with objective and quantitative judgment criteria.

[0006] The second objective of this application is to provide a diagnostic device for abnormal door closing sounds.

[0007] The third objective of this application is to propose an electronic device.

[0008] The fourth objective of this application is to provide a computer-readable storage medium.

[0009] To achieve the above objectives, the first aspect of this application proposes a method for diagnosing abnormal door closing sounds, comprising: acquiring the original audio of a vehicle door closing sound, and preprocessing the original audio to generate a target audio of the door closing sound; extracting features from the target audio to obtain acoustic features of the target audio, and generating a two-dimensional feature map based on the acoustic features; inferring from the two-dimensional feature map based on a door closing sound anomaly diagnosis model to generate inference probability data and graded quantization data; and generating anomaly diagnosis results and fault tracing suggestions based on the inference probability data and graded quantization data.

[0010] In addition, the abnormal door closing sound diagnosis method according to the above embodiments of this application may also have the following additional technical features: According to one embodiment of this application, preprocessing the original audio of a door closing sound to generate a target audio of a door closing sound includes: unifying the format of the original audio of the door closing sound to generate a standard audio of a door closing sound; denoising the standard audio of the door closing sound to generate a denoised frequency of the door closing sound; normalizing the denoised frequency of the door closing sound to generate a normalized audio of the door closing sound; and discretizing the normalized audio of the door closing sound into frames to generate the target audio of the door closing sound.

[0011] According to one embodiment of this application, the acoustic features include time-domain features, frequency-domain features, time-frequency-domain features, and fault-specific acoustic features. The generation of a two-dimensional feature map based on the acoustic features includes: normalizing the time-domain features, frequency-domain features, and fault-specific acoustic features respectively to obtain target time-domain features, target frequency-domain features, and target fault-specific acoustic features; and fusing the target time-domain features, target frequency-domain features, and target fault-specific acoustic features based on the time-frequency-domain features to generate a two-dimensional feature map.

[0012] According to one embodiment of this application, before reasoning about the two-dimensional feature map based on the door closing sound anomaly diagnostic model, the method further includes: acquiring a door closing sound sample dataset and dividing the door closing sound sample dataset to generate an initial door closing sound sample dataset; performing data augmentation on the initial door closing sound sample dataset to generate a target door closing sound sample dataset; acquiring a feature training model and setting the training parameters of the feature training model to generate an initial training model; and training and optimizing the initial training model based on the target door closing sound sample dataset to obtain the door closing sound anomaly diagnostic model.

[0013] According to one embodiment of this application, generating anomaly diagnosis results and fault tracing suggestions based on inference probability data and graded quantization data includes: in response to the inference probability data meeting preset fault conditions, determining the final fault type based on graded quantization data; generating the final fault level based on graded quantization data and preset grade thresholds; generating anomaly diagnosis results based on the final fault type and the final fault level; and obtaining fault tracing suggestions from a fault feature library based on the final fault type.

[0014] According to one embodiment of this application, in response to the inference probability data satisfying the preset fault conditions, the method further includes: determining that the original audio of the door closing sound is abnormal, and issuing an abnormality warning signal.

[0015] According to one embodiment of this application, the abnormal door closing sound diagnosis model includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; wherein, the input layer is used to receive a two-dimensional feature map; the convolutional layer is used to extract local features from the two-dimensional feature map to generate multi-channel features; the pooling layer is used to perform downsampling operations on the multi-channel features to generate dimensionality-reduced features; the fully connected layer is used to perform nonlinear mapping on the dimensionality-reduced features to generate feature vectors; and the output layer is used to output inference probability data and graded quantization data based on the feature vectors.

[0016] To achieve the above objectives, a second aspect of this application proposes a vehicle door closing sound anomaly diagnostic device, comprising: an acquisition module for acquiring the original audio of a vehicle door closing sound and preprocessing the original audio to generate a target audio of a door closing sound; a feature extraction module for extracting features from the target audio of the door closing sound to obtain acoustic features of the target audio of the door closing sound and generating a two-dimensional feature map based on the acoustic features; an inference module for inferring the two-dimensional feature map based on a door closing sound anomaly diagnostic model to generate inference probability data and grade quantization data; and a generation module for generating anomaly diagnostic results and fault tracing suggestions based on the inference probability data and grade quantization data.

[0017] The door closing sound anomaly diagnostic device of this application realizes the aforementioned door closing sound anomaly diagnostic method. This application automates and standardizes the entire door closing sound diagnostic process, effectively reducing subjective errors and labor costs caused by manual listening, and significantly improving diagnostic efficiency and consistency. Simultaneously, through comprehensive extraction and transformation of acoustic features, it enhances the ability of subsequent diagnostic stages to identify various anomalies, especially subtle abnormal noises, ultimately providing the production line with objective and quantitative judgment criteria.

[0018] To achieve the above objectives, a third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned method for diagnosing abnormal door closing sounds.

[0019] According to the electronic device of this application, the processor executes a computer program stored in the memory to implement the aforementioned method for diagnosing abnormal door closing sounds. This application automates and standardizes the entire process of diagnosing door closing sounds, effectively reducing subjective errors and labor costs caused by manual listening, and significantly improving diagnostic efficiency and consistency. Simultaneously, through comprehensive extraction and transformation of acoustic features, it enhances the ability of subsequent diagnostic stages to identify various abnormalities, especially subtle noises, ultimately providing the production line with objective and quantitative judgment criteria.

[0020] To achieve the above objectives, a fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the aforementioned method for diagnosing abnormal door closing sounds.

[0021] According to the computer-readable storage medium of this application, the aforementioned method for diagnosing abnormal door closing sounds is implemented when a computer program stored thereon is executed. This application automates and standardizes the entire process of diagnosing door closing sounds, effectively reducing subjective errors and labor costs caused by manual listening, and significantly improving diagnostic efficiency and consistency. Simultaneously, through comprehensive extraction and transformation of acoustic features, it enhances the ability of subsequent diagnostic stages to identify various abnormalities, especially subtle noises, ultimately providing the production line with objective and quantitative judgment criteria. Attached Figure Description

[0022] Figure 1 This is a flowchart of a method for diagnosing abnormal door closing sound according to some embodiments of this application; Figure 2 This is a block diagram of a vehicle door closing sound abnormality diagnostic device according to some embodiments of this application; Figure 3 This is a block diagram of an electronic device according to some embodiments of this application. Detailed Implementation

[0023] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0024] The following describes in detail, with reference to the accompanying drawings, a method, apparatus, electronic device, and storage medium for diagnosing abnormal door closing sounds according to embodiments of this application.

[0025] Figure 1 This is a flowchart of a method for diagnosing abnormal door closing sound according to some embodiments of this application. (Refer to...) Figure 1 Diagnostic methods for abnormal door closing sounds may include: S1: Acquire the original audio of the vehicle door closing sound, and preprocess the original audio of the door closing sound to generate the target audio of the door closing sound.

[0026] The raw audio of a vehicle door closing can refer to the initial digital audio signal directly captured by sound acquisition equipment (such as a high dynamic range microphone) deployed at the inspection station of the automotive production line or near the vehicle door assembly, without any modification or conversion. This signal contains the complete acoustic events generated during the door closing process, such as impact sounds, vibration sounds, locking sounds, and various potential abnormal friction or resonance sounds. It also inevitably contains background noise from the production workshop environment, electromagnetic interference, and the background noise of the acquisition equipment itself.

[0027] To clearly illustrate the preprocessing flow of this embodiment, in some embodiments of this application, the original audio of the door closing sound is preprocessed to generate the target audio of the door closing sound, including: unifying the format of the original audio of the door closing sound to generate a standard audio of the door closing sound; denoising the standard audio of the door closing sound to generate a denoised frequency of the door closing sound; normalizing the denoised frequency of the door closing sound to generate a normalized audio of the door closing sound; and discretizing the normalized audio of the door closing sound into frames to generate the target audio of the door closing sound.

[0028] Among them, the standard audio of a door closing sound can refer to an audio signal that has been standardized in format (e.g., uniform sampling rate, number of channels, bit depth, and duration). The denoised audio of a door closing sound can refer to a clean audio signal after filtering out environmental noise and electromagnetic interference. The normalized audio of a door closing sound can refer to an audio signal whose amplitude dynamic range has been mapped to a preset interval after denoising. The target audio of a door closing sound can refer to discrete audio data after dividing a continuous normalized audio signal into multiple short-time stationary frame sequences.

[0029] As a specific embodiment of this example, the original audio of the door closing sound is first standardized in format, for example, by forcibly setting it to single channel, 48 kHz sampling rate, 16-bit precision, and 1 second duration. The 1-second duration can fully cover the three stages of impact, vibration and attenuation in the door closing process to ensure that no key acoustic features are lost. If the original signal format is not compatible, standardization is completed through resampling, truncation or zero padding. The resampling adopts linear interpolation to ensure that the signal amplitude is not distorted.

[0030] Then, the standard audio of the door closing sound is decomposed into three levels using the db4 wavelet basis, separating the signal into approximate and detail components. The approximate component corresponds to the low-frequency part and represents the main acoustic characteristics of the door closing sound, while the detail component corresponds to the high-frequency part and represents the environmental noise and electromagnetic interference of the production line. A soft threshold coefficient of 0.05 is set to threshold shrink the detail component to filter out noise components. The denoised audio signal is obtained through wavelet reconstruction. In some embodiments of this application, wavelet packet decomposition can also be used for denoising, and the specific method is not limited.

[0031] Next, the noise reduction frequency amplitude of the door closing sound is mapped to the [-1, 1] interval, and the maximum amplitude normalization algorithm is used. The calculation formula can be expressed as:

[0032] Where x is the original amplitude. / The maximum / minimum amplitude of the signal. This is the normalized amplitude.

[0033] Finally, the normalized audio of the door closing sound is discretized and framed. A frame length of 20 milliseconds corresponds to 960 sampling points, a frame shift of 10 milliseconds corresponds to 480 sampling points, and a frame overlap rate of 50%. To suppress spectral leakage, a Hanning window is added to each frame. The window function formula can be expressed as: 1.

[0034] Where N is the frame length and n is the sampling point number within the frame, the output is a frame sequence composed of multiple windowed frame signals (i.e., the target audio of the door closing sound) as the standard input of the subsequent feature extraction module.

[0035] This application embodiment converts the original door closing sound audio into target audio with standardized format, noise suppression, uniform amplitude, and short-term stationary characteristics through format unification, wavelet denoising, amplitude normalization, and frame-by-frame windowing processing, providing a high-quality data input foundation for subsequent feature extraction and model diagnosis.

[0036] S2, extract features from the target audio of the door closing sound to obtain the acoustic features of the target audio of the door closing sound, and generate a two-dimensional feature map based on the acoustic features.

[0037] In some embodiments of this application, the acoustic features include time-domain features, frequency-domain features, time-frequency-domain features, and fault-specific acoustic features. The generation of a two-dimensional feature map based on the acoustic features includes: normalizing the time-domain features, frequency-domain features, and fault-specific acoustic features respectively to obtain target time-domain features, target frequency-domain features, and target fault-specific acoustic features; and fusing the target time-domain features, target frequency-domain features, and target fault-specific acoustic features based on the time-frequency-domain features to generate a two-dimensional feature map.

[0038] Temporal features refer to the statistical and morphological features extracted directly from the temporal waveform of the audio signal. They are used to describe the variation of signal amplitude over time, including the mean amplitude, peak amplitude, peak factor, pulse duration, temporal envelope slope, abnormal pulse interval, signal energy, and the proportion of low-frequency energy. To directly reflect the temporal characterization of the fault, all temporal features are calculated using intra-frame values ​​and full signal statistics (e.g., mean, variance, and maximum value).

[0039] Frequency domain features refer to the spectral distribution and morphological characteristics used to describe the energy distribution of a signal at different frequencies, including center frequency, resonant peak value, high-frequency energy proportion, frequency centroid, spectral flatness, harmonic distortion, and the number of frequency domain abrupt changes. For example, the above frequency domain features and some time-frequency domain features can be calculated after converting the framed signal to the frequency domain using a Fast Fourier Transform (FFT). The FFT point count is set to 2048, which ensures that the frequency domain resolution meets the detection requirements for minor abnormal noises.

[0040] Time-frequency domain features refer to the characteristics that describe the signal energy distribution simultaneously in both time and frequency dimensions. They are used to characterize the dynamic properties of the signal spectrum as it evolves over time, including Mel frequency cepstral coefficients and their first and second-order differences, Mel spectral coefficients, spectrogram frequency resolution, and time-frequency domain energy entropy. For example, the Mel scale number of the Mel spectrum is set to 64, and the frequency range is set to 20Hz to 20kHz. This frequency range perfectly matches the frequency response range of a car door closing sound, ensuring that the spectral features completely cover the main information and abnormal components of the door closing sound.

[0041] Fault-specific acoustic features refer to customized features specifically designed for common fault types of car door closing sounds. These features enhance the model's sensitivity to specific fault modes and include the low-frequency intermittent pulse period coefficient for characterizing hinge jamming, the correlation coefficient of high-frequency continuous scraping signals for characterizing poor seal assembly, low-frequency duration, mid-frequency resonance duration, number of high-frequency abnormal noise pulses, and frequency change amplitude. These fault-specific acoustic features are exclusive features specifically designed for common car door faults (including hinge jamming, poor seal assembly, and door lock malfunctions). They are extracted through continuous time-domain detection combined with frequency-domain threshold determination and are a key technical means to achieve accurate detection of minor abnormal noises.

[0042] Further, as a specific embodiment of this example, during the feature fusion and two-dimensional feature map construction process, a 64-dimensional Mel spectrum is used as the base time-frequency map. The normalized values ​​of the aforementioned time-domain features, frequency-domain features, and fault-specific features are mapped frame by frame to the corresponding time-frequency positions of the Mel spectrum, generating a two-dimensional Mel spectrogram feature map with a pixel size of 256×256. The mapping process employs a pixel value weighted correction method, using fault-specific features as weight coefficients to weight and enhance the pixel values ​​at corresponding positions in the map, thereby strengthening the visual representation of fault features. For example, the pulse feature of slight abnormal noise will be highlighted in the map with weight, making it easier for the convolutional neural network model to quickly capture fault features. It should be noted that all extracted features are normalized within the range of 0 to 1 to avoid model learning bias caused by differences in the units of measurement between different features.

[0043] In some alternative embodiments of this application, a logarithmic Mel energy spectrum may be used instead of the Mel spectrum in the foregoing embodiments as the base time-frequency map.

[0044] In some alternative implementations of this application, a feature stitching method can be used to replace the pixel weighting correction method in the foregoing embodiments for multi-dimensional feature fusion.

[0045] By using the Mel spectrum as a base map and normalizing multi-dimensional features before mapping them frame by frame, a standardized two-dimensional feature map adapted to the input of a convolutional neural network was generated. At the same time, fault-specific features were used as weight coefficients to weight and correct the map, which effectively enhanced the visual representation of fault-related areas. This enabled the model to capture the weak acoustic features of faults such as hinge jamming and poor sealing strip assembly more quickly and accurately, thereby significantly improving the detection sensitivity and diagnostic accuracy of minor abnormal noises.

[0046] S3 uses a door closing sound anomaly diagnostic model to infer two-dimensional feature maps to generate inference probability data and level quantization data.

[0047] In some embodiments of this application, before reasoning about the two-dimensional feature map based on the door closing sound anomaly diagnostic model, the method further includes: acquiring a door closing sound sample dataset and dividing the door closing sound sample dataset to generate an initial door closing sound sample dataset; performing data augmentation on the initial door closing sound sample dataset to generate a target door closing sound sample dataset; acquiring a feature training model and setting the training parameters of the feature training model to generate an initial training model; and training and optimizing the initial training model based on the target door closing sound sample dataset to obtain the door closing sound anomaly diagnostic model.

[0048] The abnormal door closing sound diagnostic model refers to a computational model based on a convolutional neural network, used for deep learning and inference of a two-dimensional feature map of the target audio of a door closing sound. Its input is the two-dimensional feature map generated in the above steps, and its output includes the fault type identification result and the fault level quantification value. This model undergoes supervised training using a large number of pre-labeled door closing sound samples to learn and establish a nonlinear mapping relationship between acoustic features in the two-dimensional feature map and the door fault type and fault level, thereby achieving automated and high-precision diagnosis of abnormal door closing sounds.

[0049] Furthermore, the abnormal door closing sound diagnostic model is trained in a supervised manner based on a labeled dataset of car door closing sound samples.

[0050] The door closing sound sample dataset can refer to the set of labeled audio samples used to train the door closing sound anomaly diagnosis model.

[0051] The initial door closing sound sample dataset can refer to the original sample set collected directly from the production line without any augmentation or preprocessing. Its sample size and sample diversity are limited by the collection cycle and collection conditions, and it is usually used as the basic source for data augmentation and augmentation processing.

[0052] The target door closing sound sample dataset can refer to the augmented sample set obtained after data augmentation processing of the initial door closing sound sample dataset, which is used for the final training model. This dataset is augmented through augmentation operations such as random noise addition, time-domain stretching, frequency-domain shifting, and random flipping, which significantly expands the sample size and improves sample diversity.

[0053] A feature training model can refer to a computational model framework in which two-dimensional feature maps are used as inputs and model parameters are used as learnable variables during model training, and the parameters are iteratively optimized through forward and backward propagation.

[0054] The initial training model can refer to the original model state before it has undergone sufficient iterative training, with parameters in a randomly initialized state or before fine-tuning based on pre-trained weights.

[0055] As a specific embodiment of this example, firstly, a dataset for model training is constructed. Door closing sound samples are collected from different types of car doors (e.g., sedan doors, SUV doors) and under different testing environments (e.g., different noise levels in the workshop) on the automotive production line. The dataset is constructed with a normal sample to abnormal sample ratio of 1:4. The abnormal samples cover multiple fault types (e.g., hinge jamming, poor sealing strip assembly, abnormal door lock engagement clearance, sheet metal resonance, loose interior parts) and multiple fault levels (e.g., slight abnormal noise, moderate abnormal noise, severe abnormal noise). The total sample size is no less than 100,000 samples, and the dataset is divided into a training set, a validation set, and a test set according to a preset ratio.

[0056] Secondly, by randomly adding noise (e.g., Gaussian white noise, with a signal-to-noise ratio of 20-30dB), stretching in the time domain (0.8-1.2 times), shifting in the frequency domain (±100Hz), and randomly flipping the two-dimensional feature maps in the training set of car door closing sound samples, the sample size is expanded to 500,000 samples to improve the model's generalization adaptability to different test environments and car door types.

[0057] Furthermore, an adaptive moment estimation optimizer is adopted, with a preset learning rate (e.g., 0.001) and weight decay coefficient (e.g., 1e-5). The loss function adopts a joint loss form of classification loss and regression loss, where classification loss is used to constrain the accuracy of fault type identification and regression loss is used to constrain the accuracy of fault level quantification. At the same time, the training batch size (e.g., 64) and the number of training rounds (e.g., 100) are set, and an early stopping method is adopted (e.g., if the validation set loss does not decrease for 10 consecutive rounds, training is stopped) to prevent the model from overfitting.

[0058] Finally, after the model training is completed, model quantization and layer fusion techniques are used to compress and optimize the model, remove redundant parameters, control the model size within a preset range (e.g., less than or equal to 50MB), and ensure that the model's diagnostic accuracy does not decrease significantly.

[0059] Through the synergistic effect of the above-mentioned dataset construction, data augmentation, training parameter settings, and lightweight model optimization, a door closing sound anomaly diagnostic model with high diagnostic accuracy, strong generalization ability, and lightweight deployment characteristics is finally obtained.

[0060] Among them, inference probability data refers to the classification prediction results output by the model, which is used to characterize the probability distribution of the input audio belonging to each category (including normal state and various fault types). Rank quantification data refers to the continuous value prediction results output by the model, which is used to characterize the quantification index of the severity of the fault corresponding to the input audio.

[0061] In some embodiments of this application, the door closing sound anomaly diagnosis model includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input layer receives a two-dimensional feature map; the convolutional layer extracts local features from the two-dimensional feature map to generate multi-channel features; the pooling layer performs downsampling on the multi-channel features to generate dimensionality-reduced features; the fully connected layer performs non-linear mapping on the dimensionality-reduced features to generate feature vectors; and the output layer outputs inference probability data and graded quantization data based on the feature vectors.

[0062] As a specific embodiment of this example, the door closing sound anomaly diagnosis model adopts an end-to-end convolutional neural network architecture. The overall architecture consists of a sequential connection structure from the input layer, convolutional layer, pooling layer, fully connected layer to the output layer, with no redundant network layers to adapt to the computing power of the production line computing platform. All convolutional layers use two-dimensional convolution, and the activation function uses ReLU (Rectified Linear Unit) to solve the gradient vanishing problem. The pooling layer uses max pooling to retain local maxima of features and enhance fault features. At the same time, a Dropout layer is added to prevent the model from overfitting. The specific layer design and parameters of the model are as follows: The input dimension of the input layer is 256×256×1, which is a single-channel two-dimensional Melania spectrogram feature map, used to receive the pixel matrix data of the feature map.

[0063] The first convolutional pooling block consists of: a two-dimensional convolutional layer with 32 kernels, a kernel size of 3×3, a stride of 1, and the same padding method; followed by a batch normalization layer to accelerate model convergence; then a ReLU activation layer; and finally a max pooling layer with a pooling kernel size of 2×2 and a stride of 2.

[0064] The second convolutional pooling block includes: a two-dimensional convolutional layer with 64 kernels, a kernel size of 3×3, a stride of 1, and a padding method of "same"; followed by a batch normalization layer; then a ReLU activation layer; and finally a max pooling layer with a kernel size of 2×2 and a stride of 2.

[0065] The third convolutional pooling block includes: a two-dimensional convolutional layer with 128 kernels, a kernel size of 3×3, a stride of 1, and a padding method of "same"; followed by a batch normalization layer; then a ReLU activation layer; and finally a max pooling layer with a pooling kernel size of 2×2 and a stride of 2.

[0066] The fully connected layer consists of: a flattening layer, which flattens the 3D convolutional features into a 1D vector; followed by a fully connected layer with 512 neurons; then a ReLU activation layer; and finally a Dropout layer with a dropout rate of 0.2 to prevent overfitting.

[0067] The joint output layer is divided into two branches to achieve a joint output of multi-class classification and regression. The first branch is the fault type classification branch, which contains a fully connected layer with N+1 neurons, where N is the number of common fault types of car doors, and the increment of 1 corresponds to normal samples. It is then connected to a Softmax activation function, and the output is the probability value of each fault type and normal sample, with the sum of the probability values ​​being 1. The second branch is the fault level regression branch, which contains a fully connected layer with 1 neuron, and then connected to a Sigmoid activation function. The output is a continuous quantized value in the range of 0 to 1, which corresponds to the fault level. Slight abnormal noise, moderate abnormal noise, and severe abnormal noise are mapped to different segments of the quantized value range, respectively.

[0068] In this embodiment, N is set to 5 based on common door faults on automotive production lines, specifically including: hinge jamming, poor sealing strip assembly, abnormal door lock engagement clearance, sheet metal resonance, and loose interior parts. It should be noted that those skilled in the art can adjust the value of N and the corresponding fault types according to actual application requirements.

[0069] In some alternative implementations of this application, a lightweight network architecture can be used to replace the self-developed convolutional neural network model in the foregoing embodiments.

[0070] It should be noted that, in some embodiments of this application, the TensorRT (Tensor Runtime) inference engine is used to accelerate the door closing sound anomaly diagnosis model in order to improve model inference efficiency. During inference, the two-dimensional feature map is input into the model, and after forward propagation, the fault type probability matrix and fault level quantification value are output. The inference time for a single sample model is controlled within 50 milliseconds. In some alternative embodiments of this application, the ONNXRuntime (Open Neural Network Exchange Runtime) inference engine can be used to replace the TensorRT engine in the aforementioned embodiments for model acceleration.

[0071] S4 generates anomaly diagnosis results and fault tracing suggestions based on inference probability data and graded quantification data.

[0072] In some embodiments of this application, generating anomaly diagnosis results and fault tracing suggestions based on inference probability data and grading quantization data includes: in response to the inference probability data meeting preset fault conditions, determining the final fault type based on grading quantization data; generating the final fault level based on grading quantization data and preset grading threshold; generating anomaly diagnosis results based on the final fault type and the final fault level; and obtaining fault tracing suggestions from a fault feature library according to the final fault type.

[0073] The preset fault condition can refer to the judgment rule used to determine whether the audio sample is abnormal and to determine the specific fault type. For example, the preset fault condition can be set to determine that the sample is abnormal and the fault type is identified as a candidate fault type when the probability value of a certain fault type in the inference probability data reaches or exceeds a preset probability threshold (e.g., 0.8).

[0074] Specifically, when the probability values ​​of multiple fault types in the inference probability data all meet the preset fault conditions, the fault type with the highest probability value is selected as the final fault type; when only one fault type meets the conditions, that fault type is directly determined as the final fault type; when the probability values ​​of all fault types do not meet the preset fault conditions, the sample is determined to be in a normal state.

[0075] In some embodiments, when the probability values ​​of all fault types in the inference probability data do not meet the preset fault conditions (i.e., all are below the preset probability threshold), the system determines that the current door closing sound sample is in a normal state, does not perform abnormal warning and fault tracing operations, and outputs a normal diagnostic result.

[0076] When the probability values ​​of multiple fault types in the inference probability data simultaneously meet the preset fault conditions (i.e., all reach or exceed the preset probability threshold), the system determines that the current door closing sound sample has multiple anomalies. In this case, the fault type with the highest probability value is selected as the primary fault type, and the remaining fault types that meet the conditions are recorded as accompanying fault types in the anomaly diagnosis results, thereby achieving joint diagnosis and output of multiple faults.

[0077] The preset level threshold can refer to the boundary value used to divide continuous level quantized data into discrete fault levels. For example, the preset level threshold can be set to two boundary values ​​of 0.3 and 0.7, dividing the quantization interval from 0 to 1 into three sub-intervals, corresponding to three fault levels: slight abnormal noise, moderate abnormal noise, and severe abnormal noise, respectively.

[0078] As a specific implementation, the level quantization value output by the model is compared with the preset level threshold. The corresponding fault level is determined according to the threshold range in which the value falls. For example, when the quantization value is less than 0.3, it is judged as a slight abnormal noise; when it is between 0.3 and 0.7, it is judged as a moderate abnormal noise; and when it is greater than 0.7, it is judged as a severe abnormal noise.

[0079] It should be noted that the preset fault conditions and preset level thresholds can be set by technicians according to the actual situation, and there are no specific restrictions.

[0080] Anomaly diagnosis results are generated based on the final fault type and final fault level; fault tracing suggestions are obtained from the fault feature library based on the final fault type.

[0081] The fault feature database is a pre-built relational database that stores the mapping relationship between each fault type and its corresponding traceability information. The traceability information includes a description of the fault cause, the location of the faulty component, and rework operation suggestions. By querying the fault feature database, traceability suggestions for specific fault types can be automatically generated, providing direct guidance for rework operations on the production line.

[0082] In some embodiments of this application, in response to the inference probability data meeting the preset fault conditions, the method further includes: determining that the original audio of the door closing sound is abnormal, and issuing an abnormal warning signal.

[0083] Specifically, when the probability value of any fault type in the inference probability data output by the model reaches or exceeds the probability threshold set by the preset fault conditions, the system determines that the currently detected door closing sound is abnormal. In response to this abnormality determination, the system automatically generates an abnormality warning signal. This warning signal can be sent to the production line control system in the form of a digital level signal (e.g., high level 1) to trigger the on-site audible and visual alarm device, reminding production line staff to pay attention to the abnormal door and take subsequent actions. Through this abnormality warning mechanism, real-time feedback of the abnormal door closing sound detection results and the coordinated response of the production line are achieved, effectively improving the efficiency of abnormality handling in the production line's quality control process.

[0084] In some embodiments of this application, the entire process data for each vehicle door sample is fully stored. The stored data types include raw audio signals, preprocessing data, two-dimensional feature maps, model inference results, and final diagnostic results. The storage format combines CSV (Comma-Separated Values) and WAV (Waveform Audio File Format), with structured data stored in CSV format and audio and feature map data stored in WAV format. The storage system supports retrieval and querying based on multiple dimensions such as inspection time, vehicle door model, and workstation number. Simultaneously, the stored data can be seamlessly integrated with the vehicle manufacturer's quality control system, providing complete data support for subsequent quality traceability, data analysis, and model iteration.

[0085] The embodiments of this application achieve automated determination of the presence, type, and level of anomalies by using a hierarchical judgment based on preset fault conditions and level thresholds. For multiple anomaly scenarios, a method of jointly outputting the main fault with the most probable accompanying fault is adopted to avoid information loss. By automatically obtaining tracing suggestions from the fault feature library, direct guidance is provided for production line rework. At the same time, by sending anomaly warning signals in real time, linkage response with the production line is realized, effectively improving the efficiency of anomaly handling in the quality control process.

[0086] In one specific embodiment of this application, the software solution of the above-mentioned abnormal door closing sound diagnosis system is deployed on an embedded industrial control computer at the production line inspection station. The hardware configuration of the industrial control computer is as follows: the CPU (Central Processing Unit) is a quad-core Cortex-A72 processor with an ARM (Advanced RISC Machines) architecture, with a main frequency of 2.0GHz; the memory capacity is 8GB; and the storage uses 64GB of eMMC (Embedded MultiMedia Card) flash memory. The operating system is Ubuntu 20.04 LTS (Ubuntu 20.04 Long Term Support). The software runtime environment relies on the following core libraries: Python 3.8 (Python 3.8, Python programming language version 3.8) as the primary programming language, Librosa 0.10.1 (Librosa 0.10.1, audio processing library) for audio processing, TensorFlow 2.15.0 (TensorFlow 2.15.0, TensorFlow) for model training, TensorRT 8.6.1 (Tensor Runtime 8.6.1, TensorRuntime) for accelerating model inference, and OpenCV 4.8.0 (Open Source Computer Vision Library 4.8.0, open source computer vision library) for feature map processing.

[0087] It should be noted that the above hardware configuration and software version are merely specific examples and are not intended to limit the scope of protection of this application. In alternative implementations, the operating system may be other distributions of Linux or Windows. The deep learning framework may be TensorFlow 2.x or PyTorch (Python Torch) for building, training, and deploying multi-task deep learning models. Librosa (Librosa, an audio processing library) is used for audio file loading, normalization, and advanced acoustic feature extraction. Scikit-learn (Scikit-learn, a machine learning toolkit) is used for data partitioning and performance metric calculation, including accuracy and confusion matrices. NumPy (NumericalPython) and Pandas (Panel Data) are used for efficient numerical computation and data processing. Matplotlib (Mathematical Plotting Library) and Seaborn (Seaborn, a marine statistical plotting library) are used to generate various visualization charts, such as loss curves, confusion matrices, and scatter plots. Those skilled in the art can flexibly select the above-mentioned hardware and software configurations according to the actual deployment environment and performance requirements.

[0088] The method for diagnosing abnormal door closing sounds in this application has the following technical advantages compared to related technologies: 1. Significantly improved diagnostic accuracy, effectively avoiding missed detection of minor abnormal noises: By extracting time-domain, frequency-domain, time-frequency-domain, and fault-specific acoustic features and fusing them to generate a two-dimensional feature map, the shortcomings of traditional methods, such as single features and missing key information, are overcome. In particular, the dedicated features designed for hinge jamming, poor sealing strip assembly, etc., enable the model to accurately capture the faint acoustic characteristics of minor abnormal noises, greatly reducing the rate of missed and false detections.

[0089] 2. Significantly improved diagnostic efficiency, fully adaptable to the batch testing needs of the production line: The automated diagnostic process can realize real-time acquisition and rapid processing of door closing sound samples. The overall diagnostic time for a single sample is controlled within 110 milliseconds, which is far higher than the efficiency of manual diagnosis. It fully meets the batch testing requirements of the automotive production line to process 2 to 3 door samples per minute, breaking through the technical problem that manual diagnosis has become a bottleneck in quality control efficiency.

[0090] 3. Integrated detection for anomaly diagnosis and fault tracing: Through the model's multi-task joint output architecture, it simultaneously outputs fault type identification results and fault level quantification values, overcoming the limitation of traditional technologies that only output "normal / abnormal" binary classification results. Combined with a fault feature library, it automatically provides fault causes, component location, and repair suggestions, eliminating the need for subsequent manual analysis and achieving closed-loop management from anomaly detection to fault root cause location.

[0091] 4. Strong model generalization ability and excellent scenario adaptability: By employing convolutional neural networks combined with data augmentation techniques, the model possesses strong adaptability to different door types (car doors, SUV doors), different testing environments (changes in workshop background noise, temperature and humidity fluctuations), and novel fault characteristics. After scenario changes, there is no need to re-label a large number of samples; only minor fine-tuning is required to maintain diagnostic accuracy, resulting in low model iteration costs.

[0092] 5. Significantly reduces quality control labor costs and improves the technological adaptability of small and medium-sized car manufacturers: There is no need to train and maintain a professional team for manually diagnosing door noises; only a small number of staff are required to operate and maintain the audio acquisition equipment, greatly saving on labor training and management costs. The system is easy to operate and has a low technical threshold, effectively solving the problem that small and medium-sized car manufacturers cannot achieve high-precision door noise diagnosis due to cost constraints.

[0093] Corresponding to the above embodiments, this application also provides a diagnostic device for abnormal door closing sound, referring to... Figure 2 The device 200 includes: an acquisition module 210, a feature extraction module 220, an inference module 230, and a generation module 240.

[0094] The acquisition module 210 is used to acquire the original audio of the vehicle door closing sound and preprocess the original audio to generate the target audio of the door closing sound; the feature extraction module 220 is used to extract features from the target audio of the door closing sound to obtain the acoustic features of the target audio of the door closing sound and generate a two-dimensional feature map based on the acoustic features; the inference module 230 is used to infer the two-dimensional feature map based on the door closing sound anomaly diagnosis model to generate inference probability data and level quantization data; the generation module 240 is used to generate anomaly diagnosis results and fault tracing suggestions based on the inference probability data and level quantization data.

[0095] In some embodiments of this application, the acquisition module 210 preprocesses the original audio of the door closing sound to generate the target audio of the door closing sound. Specifically, it is used to: unify the format of the original audio of the door closing sound to generate the standard audio of the door closing sound; denoise the standard audio of the door closing sound to generate the denoised audio of the door closing sound; normalize the denoised audio of the door closing sound to generate the normalized audio of the door closing sound; and discretize and frame the normalized audio of the door closing sound to generate the target audio of the door closing sound.

[0096] In some embodiments of this application, the feature extraction module 220 determines that the acoustic features include time-domain features, frequency-domain features, time-frequency-domain features, and fault-specific acoustic features. Specifically, generating a two-dimensional feature map based on the acoustic features involves: normalizing the time-domain features, frequency-domain features, and fault-specific acoustic features respectively to obtain target time-domain features, target frequency-domain features, and target fault-specific acoustic features; and fusing the target time-domain features, target frequency-domain features, and target fault-specific acoustic features based on the time-frequency-domain features to generate a two-dimensional feature map.

[0097] In some embodiments of this application, before the inference module 230 performs inference on the two-dimensional feature map based on the door closing sound anomaly diagnostic model, it is further configured to: acquire a door closing sound sample dataset and divide the door closing sound sample dataset to generate an initial door closing sound sample dataset; perform data augmentation on the initial door closing sound sample dataset to generate a target door closing sound sample dataset; acquire a feature training model and set the training parameters of the feature training model to generate an initial training model; and train and optimize the initial training model based on the target door closing sound sample dataset to obtain the door closing sound anomaly diagnostic model.

[0098] In some embodiments of this application, the generation module 240 generates anomaly diagnosis results and fault tracing suggestions based on inference probability data and grading quantization data. Specifically, it is used to: determine the final fault type based on grading quantization data in response to the inference probability data meeting preset fault conditions; generate the final fault level based on grading quantization data and preset grading threshold; generate anomaly diagnosis results based on the final fault type and the final fault level; and obtain fault tracing suggestions from the fault feature library according to the final fault type.

[0099] In some embodiments of this application, after the generation module 240 responds to the inference probability data meeting the preset fault conditions, it is further configured to: determine that the original audio of the door closing sound is abnormal and issue an abnormal warning signal.

[0100] In some embodiments of this application, the inference module 230 determines that the abnormal door closing sound diagnostic model includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input layer receives a two-dimensional feature map; the convolutional layer extracts local features from the two-dimensional feature map to generate multi-channel features; the pooling layer performs downsampling on the multi-channel features to generate dimensionality-reduced features; the fully connected layer performs non-linear mapping on the dimensionality-reduced features to generate feature vectors; and the output layer outputs inference probability data and graded quantization data based on the feature vectors.

[0101] It should be noted that for details not disclosed in the diagnostic device for abnormal door closing sound in this application, please refer to the details disclosed in the diagnostic method for abnormal door closing sound, and no specific limitations are imposed.

[0102] Corresponding to the above embodiments, this application also provides an electronic device, specifically referring to... Figure 3 The electronic device 300 includes: a memory 310, a processor 320, and a computer program stored on the memory 310 and executable on the processor 320. The processor 320 executes the program to implement the aforementioned method for diagnosing abnormal door closing sounds.

[0103] This application also provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the aforementioned method for diagnosing abnormal door closing sounds.

[0104] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0105] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0106] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0107] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

[0108] Any process or method described in the flowchart or otherwise herein is to be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0109] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0110] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0111] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0112] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0113] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

[0114] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this application can be achieved, and this is not limited herein.

[0115] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for diagnosing abnormal door closing sound, characterized in that, include: The original audio of the vehicle door closing sound is acquired, and the original audio of the door closing sound is preprocessed to generate the target audio of the door closing sound. Feature extraction is performed on the target audio of the door closing sound to obtain its acoustic features, and a two-dimensional feature map is generated based on the acoustic features; The two-dimensional feature map is inferred based on the abnormal door closing sound diagnosis model to generate inference probability data and level quantization data. Anomaly diagnosis results and fault tracing suggestions are generated based on the inference probability data and the level quantification data.

2. The method according to claim 1, characterized in that, The preprocessing of the original audio of the door closing sound to generate the target audio of the door closing sound includes: The original audio of the door closing sound is formatted to generate a standard audio of the door closing sound; The standard audio of the door closing sound is denoised to generate a denoised door closing sound frequency. The noise-reduced frequency of the door closing sound is normalized to generate a normalized audio of the door closing sound. The normalized audio of the door closing sound is discretized and framed to generate the target audio of the door closing sound.

3. The method according to claim 1, characterized in that, The acoustic features include time-domain features, frequency-domain features, time-frequency-domain features, and fault-specific acoustic features, wherein generating a two-dimensional feature map based on the acoustic features includes: The time-domain features, frequency-domain features, and fault-specific acoustic features are normalized respectively to obtain the target time-domain features, target frequency-domain features, and target fault-specific acoustic features. Based on the time-frequency domain features, the target time-domain features, the target frequency-domain features, and the target fault-specific acoustic features are fused to generate the two-dimensional feature map.

4. The method according to claim 1, characterized in that, Before inferring the two-dimensional feature map based on the door closing sound anomaly diagnostic model, the method further includes: Obtain a sample dataset of door closing sounds and divide the sample dataset to generate an initial sample dataset of door closing sounds; Data augmentation is performed on the initial door closing sound sample dataset to generate a target door closing sound sample dataset; Obtain the feature training model and set the training parameters of the feature training model to generate an initial training model; The initial training model is trained and optimized based on the target door closing sound sample dataset to obtain the door closing sound anomaly diagnosis model.

5. The method according to claim 1, characterized in that, The process of generating anomaly diagnosis results and fault tracing suggestions based on the inference probability data and the level quantification data includes: In response to the inference probability data satisfying preset fault conditions, the final fault type is determined based on the level quantification data; The final fault level is generated based on the quantified level data and the preset level threshold. The abnormal diagnosis result is generated based on the final fault type and the final fault level; Based on the final fault type, obtain the fault tracing suggestion from the fault feature library.

6. The method according to claim 5, characterized in that, The response after the inference probability data satisfies the preset fault condition further includes: The original audio of the door closing sound is determined to be abnormal, and an abnormality warning signal is issued.

7. The method according to claim 1, characterized in that, The abnormal door closing sound diagnostic model includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer, wherein... The input layer is used to receive the two-dimensional feature map; The convolutional layer is used to extract local features from the two-dimensional feature map to generate multi-channel features; The pooling layer is used to downsample the multi-channel features to generate dimensionality-reduced features; The fully connected layer is used to perform nonlinear mapping on the dimensionality-reduced features to generate feature vectors; The output layer is used to output the inference probability data and the level quantization data based on the feature vector.

8. A diagnostic device for abnormal door closing sound, characterized in that, include: The acquisition module is used to acquire the original audio of the vehicle door closing sound and preprocess the original audio of the door closing sound to generate the target audio of the door closing sound. The feature extraction module is used to extract features from the target audio of the door closing sound to obtain the acoustic features of the target audio of the door closing sound, and generate a two-dimensional feature map based on the acoustic features; The inference module is used to infer the two-dimensional feature map based on the abnormal door closing sound diagnosis model to generate inference probability data and level quantization data. The generation module is used to generate anomaly diagnosis results and fault tracing suggestions based on the inference probability data and the level quantification data.

9. An electronic device, characterized in that, include: The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the abnormal door closing sound diagnosis method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the abnormal door closing sound diagnosis method as described in any one of claims 1-7.

Images