Abnormal sound detection method and system for air conditioner outdoor unit based on generative model

Abstract

The present invention provides a generative model-based method and system for detecting abnormal sounds from an air conditioner outdoor unit. The method comprises obtaining normal operating sound data of the air conditioner outdoor unit and interference sound data from the air conditioner detection environment; preprocessing the sound data to obtain time-frequency features; training a Gaussian mixture model based on the time-frequency features of the interference sound data to obtain its data distribution, which is then used for random sampling. The sampled sound data is then superimposed on the time-frequency features of the normal air conditioner sound data to obtain fifth sound data and corresponding sampling step labels. A neural network model is trained using the normal operating sound data and the fifth sound data to obtain a trained anti-interference abnormal sound detection model for the air conditioner outdoor unit; obtaining the air conditioner operating sound data to be tested, preprocessing it, and inputting it into the trained abnormal sound detection model to determine whether the air conditioner operating sound is abnormal. The present invention improves the anti-interference performance of the detection algorithm and the efficiency of air conditioner outdoor unit detection, thereby reducing labor costs.

Description

Technical Field

[0001] The present invention relates to the field of detection technology for air conditioner outdoor units, and in particular to the fields of deep learning, machine learning, and audio processing. More specifically, the present invention provides a method and system for detecting abnormal sounds from air conditioner outdoor units based on a generative model. Background Art

[0002] Before leaving the factory, air conditioner outdoor units undergo a pre-operational test, where manual auscultation is performed to confirm the presence of abnormal noise. This method is inefficient, and human subjectivity significantly impacts the results. With the rapid development of artificial intelligence and deep learning technologies, sound anomaly detection technology based on acoustic sensing has been increasingly proposed and has achieved considerable success.

[0003] Unsupervised learning-based sound anomaly detection technology typically classifies sounds that match the learned characteristics (such as normal air conditioner operation) as normal, while classifying sounds that differ from normal (such as abnormal air conditioner operation and other non-air conditioner operation sounds that have not been learned) as abnormal. However, in real-world industrial scenarios such as air conditioner operation testing, where background noise is complex, the microphone not only picks up the air conditioner's operating sound, but also surrounding interference noise such as motor noise and valve noise. In the presence of these interferences, conventional abnormal sound detection methods cannot accurately determine whether the air conditioner's operating sound is normal. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to propose a method and system for detecting abnormal sounds of air-conditioning outdoor units based on a generative model, which can perform contactless real-time detection of the operating sounds of air-conditioning outdoor units under complex industrial noise interference conditions, thereby reducing labor costs and improving the production quality of air-conditioning outdoor units.

[0005] To achieve the above-mentioned and other related purposes, the present invention provides the following technical solutions:

[0006] In a first aspect, the present invention provides a method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model, comprising the following steps:

[0007] Acquire sound data of normal operation of the air conditioner outdoor unit to obtain first sound data, and acquire interference sound data in the environment where the air conditioner is tested to obtain second sound data;

[0008] Preprocessing the first sound data and the second sound data to obtain time-frequency features, thereby obtaining third sound data and fourth sound data;

[0009] training a Gaussian mixture model using the fourth sound data to obtain a fourth sound data distribution, and superimposing sound data sampled from the fourth sound data distribution onto the third sound data to obtain fifth sound data and a corresponding sampling step label;

[0010] Obtaining a neural network model, wherein the neural network model includes an audio noise reduction module and an anomaly detection module;

[0011] Training the neural network model using the third sound data and the fifth sound data, first fixing the weights of the audio noise reduction module, using the third sound data to train and update the parameters of the anomaly detection module, then fixing the weights of the anomaly detection module, and using the fifth sound data and sampling step labels to train the parameters of the audio noise reduction module;

[0012] According to the neural network model update method, the method is repeated multiple times until the neural network model converges, thereby obtaining a trained air conditioner outdoor unit anti-interference abnormal noise detection model;

[0013] Acquire the operating sound data of the air conditioner to be tested, and obtain sixth sound data through preprocessing, and use the trained anti-interference abnormal sound detection model of the air conditioner outdoor unit to perform abnormality detection on the sixth sound data.

[0014] In some optional implementations, obtaining interference sound data in the environment where the air conditioner is detected includes:

[0015] The multi-channel sound transmission device collects the running sound of the transmission device, the movement sound of the cylinder, the running sound of the motor and the movement and impact sound of the mechanical structure.

[0016] In some optional implementations, the preprocessing of the first sound data and the second sound data is time-frequency feature transformation, and the processing process includes:

[0017] First, the sound data is framed and windowed, with some overlap between frames. Then, a short-time Fourier transform is performed on each frame of data to obtain the logarithmic time-frequency feature map.

[0018] In some optional embodiments, the fourth sound data is used to train a Gaussian mixture model to determine the distribution of the fourth sound data. include:

[0019] Initialize the Gaussian mixture model parameters and use the EM algorithm to iteratively solve the fourth sound data distribution, its density function Expressed as:

[0020]

[0021] Where x is the sound data, M is the number of components of the Gaussian mixture model, μ i is the mean of the corresponding component, ∑i is the covariance matrix of the corresponding component, n is the dimension of sample x, i is the i-th Gaussian distribution function, and T is the matrix transpose.

[0022] In some optional embodiments, based on the third sound data having N samples, t interference noise samples are randomly sampled from the Gaussian mixture density function in t steps for each sample in the third sound data, and are superimposed one by one with the third sound data to obtain the fifth sound data containing N×t sound samples, and N×t corresponding sampling step labels S.

[0023] In some optional embodiments, the fifth sound data is obtained by combining the third sound data and the density function The noise samples obtained by sampling are superimposed and described as:

[0024]

[0025] in, is the i-th sample in the third sound data, for The fifth sound data sample is generated by superimposing the interference noise obtained by sampling in the tth step, and the weight α t =1-β t , for is the noise obtained by sampling, β is the manually set parameter, j is the jth step;

[0026] The sampling step label S is normalized to (0, 1).

[0027] In some optional implementations, the audio noise reduction module is a diffusion model, and the anomaly detection module is a deep convolutional autoencoder model.

[0028] In some optional embodiments, training the neural network model includes:

[0029] First, the third sound data is input into the network, the weight of the audio noise reduction module is fixed, and the anomaly detection module reconstructs the third sound data to update the network. Then, the weight of the anomaly detection module is fixed, and the audio noise reduction module sequentially estimates the interference noise superimposed on the fifth sound data and generates an estimated sampling step corresponding to the interference noise. The fifth sound data is then subtracted from the estimated interference noise and re-input into the audio noise reduction module. After all the fifth sound data are estimated, the fifth sound data obtained in the last step is input into the anomaly detection module. This process is repeated until the neural network model converges. The loss function of the neural network model is determined according to the following formula:

[0030]

[0031] Where ∈ is the interference noise obtained by actual sampling, ∈ θ is the interference noise estimated by the audio noise reduction module, t is the true value of the interference noise sampling step, is the interference noise sampling step prediction value estimated by the audio noise reduction module, x is the sound data after noise reduction, Reconstructed sound data for the anomaly detection module; θ and γ are both weight parameters of network loss.

[0032] In some optional implementations, the trained anti-interference abnormal sound detection model for the air conditioner outdoor unit includes:

[0033] The acquired air conditioner outdoor unit operating sound data is preprocessed to obtain time-frequency features, which are then input into the trained air conditioner outdoor unit anti-interference abnormal sound detection model; first, the estimated noise ∈ θ , and the estimated sampling step Then, the time-frequency feature is subtracted from the estimated noise and then re-input into the audio noise reduction module until the estimated sampling step output by the audio noise reduction module is If the value is less than the threshold th_denoise~(0,1), the corresponding audio noise reduction module input is subtracted from the estimated interference noise and then input into the abnormality detection module to obtain the reconstruction error e; when the reconstruction error e is greater than the threshold th, it is judged that the air conditioner operation sound is abnormal, otherwise it is normal.

[0034] In a second aspect, the present application provides an abnormal sound detection system for an air conditioner outdoor unit based on a generative model, comprising:

[0035] An audio data acquisition unit is configured to acquire sound data of normal operation of the air conditioner outdoor unit to obtain first sound data, and acquire interference sound data in the environment in which the air conditioner is tested to obtain second sound data;

[0036] a time-frequency feature extraction unit, configured to pre-process the first sound data and the second sound data to obtain time-frequency features, and then obtain third sound data and fourth sound data;

[0037] a Gaussian mixture model training unit, configured to train a Gaussian mixture model using the fourth sound data to obtain a fourth sound data distribution, and to superimpose sound data sampled from the fourth sound data distribution onto the third sound data to obtain fifth sound data and a corresponding sampling step label;

[0038] A neural network model acquisition unit, configured to acquire a neural network model, wherein the neural network model includes an audio noise reduction module and an anomaly detection module;

[0039] a neural network model training unit, configured to train the neural network model using the third sound data and the fifth sound data, first fixing the weights of the audio noise reduction module, using the third sound data to train and update the parameters of the anomaly detection module, then fixing the weights of the anomaly detection module, and training the parameters of the audio noise reduction module using the fifth sound data and sampling step labels;

[0040] an abnormal sound detection model obtaining unit, configured to obtain a trained anti-interference abnormal sound detection model for an air conditioner outdoor unit by repeatedly updating the neural network model until the neural network model converges;

[0041] The abnormal sound detection unit is used to obtain the operating sound data of the air conditioner to be tested, obtain sixth sound data through preprocessing, and perform abnormality detection on the sixth sound data using the trained anti-interference abnormal sound detection model of the air conditioner outdoor unit.

[0042] The generative model-based method for detecting abnormal sound from an air conditioner outdoor unit, provided by the present invention, acquires sound data from the air conditioner's operation and, based on this sound data, determines whether the unit's operating status is normal. Compared to traditional manual auscultation, the generative model-based method for detecting abnormal sound from an air conditioner outdoor unit, provided by the present invention, enables contactless, real-time detection of the unit's operating status even under complex industrial noise interference conditions, thereby reducing the defective rate of air conditioner outdoor units during factory production and improving their production quality.

[0043] It should be understood that the abnormal sound detection system for the air conditioner outdoor unit based on the generative model provided in the second aspect corresponds to the method in the first aspect. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method provided above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0045] Figure 1 This is a flow chart of a method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model provided by an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the training process of the anti-interference abnormal noise detection model for the air conditioner outdoor unit provided by an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of the detection process of the anti-interference abnormal noise detection model for the air conditioner outdoor unit provided by an embodiment of the present invention;

[0048] Figure 4 Schematic diagram of an abnormal sound detection system for an air conditioner outdoor unit based on a generative model provided by an embodiment of the present invention;

[0049] Figure 5 This is a block diagram of an electronic device provided in an example of this application. DETAILED DESCRIPTION

[0050] For ease of understanding, some explanations of concepts related to the embodiments of the present application are exemplarily provided for reference. It should be noted that in the present application, "at least one" refers to one or more, and "a plurality" refers to two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims, and drawings of this application are used to distinguish similar objects, rather than to describe a specific order or sequence.

[0051] In order to better understand the abnormal sound detection method and system for the air conditioner outdoor unit based on the generative model provided in the embodiment of the present invention, the embodiment of the present application will be introduced below with reference to the accompanying drawings.

[0052] Example 1

[0053] Figure 1 This is a flow chart of a method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model provided by an embodiment of the present invention.

[0054] The present invention provides a method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model, comprising the following steps:

[0055] S101, obtaining sound data of normal operation of an air conditioner outdoor unit to obtain first sound data, and obtaining interference sound data in an environment where the air conditioner is detected to obtain second sound data.

[0056] The first sound data is collected by a multi-channel microphone from the air conditioner's normal operation in a static environment. The second sound data is interference sound data from the air conditioner's production test environment, such as the sound of a running conveyor belt, valve operation, mechanical mechanism operation, and motor operation.

[0057] S102 , preprocessing the first sound data and the second sound data to obtain time-frequency features, thereby obtaining third sound data and fourth sound data.

[0058] The sound features extracted from sound data may include time domain features such as short-time average energy, short-time average zero-crossing rate, resonance peak, pitch period, etc., and may also include transform domain features such as linear prediction coefficient (LPC), linear prediction cepstrum coefficient (LPCC), Mel frequency cepstrum coefficient (MFCC), and perceptual linear prediction (PLP).

[0059] In one embodiment of the present application, extracting logarithmic time-frequency features from sound data specifically includes: performing a discrete Fourier transform (DFT, such as a fast Fourier transform) on the sound data to obtain a spectrum of the sound data; calculating a discrete energy spectrum of the sound data based on the spectrum of the sound data; and performing a logarithmic operation on the discrete energy spectrum of the sound data to obtain a logarithmic power spectrum of the sound data.

[0060] In one embodiment of the present application, when performing discrete Fourier transform on the sound data, the number of Fourier transform points is 2048.

[0061] In the embodiment of the present application, the sound features corresponding to one frame of sound data can be used as a sound sample, or the sound features corresponding to multiple frames of sound data can be used as a sound sample. If the sound features corresponding to every n frames of sound data are used as a sound sample, then m sound samples can be obtained per second of sound data:

[0062]

[0063] in, Indicates rounding down, fs is the sampling frequency, n_fft is the number of Fourier transform points per frame, n is the number of frames corresponding to each sound sample, and n is an integer greater than or equal to 1.

[0064] For example, if the number of Fourier transform points per frame is 2048 and n is 5, then

[0065] In one embodiment of the present application, the sampling frequency is 44.1 kHz.

[0066] In the embodiment of the present application, preprocessing the sound data may include pre-emphasizing the sound data.

[0067] The purpose of pre-emphasis is to enhance the high-frequency components in the sound data and make the signal spectrum flat. Pre-emphasis is generally implemented using a high-pass filter, and the transfer function of the high-pass filter can be:

[0068] H(z)=1-kz -1 , 0.9≤k≤1.0.

[0069] Wherein, k is the pre-emphasis coefficient, and its preferred value is between 0.94-0.97.

[0070] Preprocessing the sound data may further include dividing the sound data into frames.

[0071] Framing is achieved by windowing the sound data. To avoid excessive changes in the amplitude of two adjacent frames, the frames need to overlap. In one embodiment of the present application, each frame is 25 milliseconds, and there is a 15 millisecond overlap between two adjacent frames, that is, a frame is taken every 10 milliseconds.

[0072] Commonly used window functions include rectangular window, Hamming window and Hanning window.

[0073] S103 : Using the fourth sound data to train a Gaussian mixture model to obtain a fourth sound data distribution, and adding the sound data sampled from the fourth sound data distribution to the third sound data to obtain fifth sound data and a corresponding sampling step label.

[0074] In one embodiment of the present application, a Gaussian mixture model is trained using the fourth sound data to obtain the probability density function of the collected interference noise distribution. It can be described as:

[0075]

[0076] Where x is the sound data, M is the number of components of the Gaussian mixture model, μ i is the mean of the corresponding component, ∑ i is the covariance matrix of the corresponding component, n is the dimension of sample x, i is the i-th Gaussian distribution function, and T is the matrix transpose.

[0077] In one example of the present application, based on the third sound data having N samples, t interference noise samples are randomly sampled from the Gaussian mixture density function in t steps for each sample in the third sound data, and are superimposed one by one with the third sound data to obtain the fifth sound data containing N×t sound samples, and N×t corresponding sampling step labels S, the sampling step t is set to 1000, the sampling step interval is 2, and β0=0.0001.

[0078] Furthermore, the fifth sound data is obtained by combining the third sound data and the density function The noise samples obtained by sampling are superimposed and described as:

[0079]

[0080] in, is the i-th sample in the third sound data, for The fifth sound data sample is generated by superimposing the interference noise obtained by sampling in the tth step, and the weight α t =1-β t , for is the noise obtained by sampling, β is the manually set parameter, j is the jth step;

[0081] The corresponding sampling step label S is normalized to (0, 1), and the label value of any step t is t / 1000.

[0082] S104: Acquire a neural network model, where the neural network model includes an audio noise reduction module and an anomaly detection module.

[0083] In one example of the present application, the audio noise reduction module is a diffusion model, whose network structure is a U-net network, and the anomaly detection module is a deep convolutional autoencoder model.

[0084] S105, training the neural network model using the third sound data and the fifth sound data, first fixing the weights of the audio noise reduction module, using the third sound data to train and update the parameters of the anomaly detection module, then fixing the weights of the anomaly detection module, and using the fifth sound data and sampling step labels to train the parameters of the audio noise reduction module.

[0085] like Figure 2 As shown, in some optional embodiments, training the neural network model includes:

[0086] First, the third sound data is input into the network, the weight of the audio noise reduction module is fixed, and the anomaly detection module reconstructs the third sound data to update the network. Then, the weight of the anomaly detection module is fixed, and the audio noise reduction module sequentially estimates the interference noise superimposed on the fifth sound data and generates an estimated sampling step corresponding to the interference noise. The fifth sound data is then subtracted from the estimated interference noise and re-input into the audio noise reduction module. After all the fifth sound data are estimated, the fifth sound data obtained in the last step is input into the anomaly detection module. This process is repeated until the neural network model converges. The loss function of the neural network model is determined according to the following formula:

[0087]

[0088] Where ∈ is the interference noise obtained by actual sampling, ∈ θ is the interference noise estimated by the audio noise reduction module, t is the true value of the interference noise sampling step, is the interference noise sampling step prediction value estimated by the audio noise reduction module, x is the sound data after noise reduction, Reconstructed sound data for the anomaly detection module. θ and γ are both weight parameters of network loss. θ=1000,γ=1.

[0089] S106, participating in an update method according to the neural network model, repeating multiple times until the neural network model converges, and obtaining a trained air conditioner outdoor unit anti-interference abnormal sound detection model.

[0090] S107, using the trained anti-interference abnormal sound detection model for the air-conditioning outdoor unit to perform abnormality detection on the operating sound of the air-conditioning outdoor unit.

[0091] like Figure 3 As shown, in some optional implementations, the trained air conditioner outdoor unit anti-interference abnormal sound detection model includes:

[0092] The acquired air conditioner outdoor unit operating sound data is preprocessed to obtain time-frequency features, which are then input into the trained air conditioner outdoor unit anti-interference abnormal sound detection model. First, the estimated noise ∈ θ , and the estimated sampling step Then, the time-frequency feature is subtracted from the estimated noise and then re-input into the audio noise reduction module until the estimated sampling step output by the audio noise reduction module is If the value is less than the threshold th_denoise~(0,1), the corresponding audio noise reduction module input is subtracted from the estimated interference noise and then input into the anomaly detection module to obtain the reconstruction error e. If the reconstruction error e is greater than the threshold th, the air conditioner is judged to be operating abnormally, otherwise it is normal.

[0093] Example 2

[0094] like Figure 4 As shown, the present application provides an abnormal sound detection system for an air conditioner outdoor unit based on a generative model, comprising an audio data acquisition unit, a time-frequency feature extraction unit, a Gaussian mixture model training unit, a neural network model acquisition unit, a neural network model training unit, an abnormal sound detection model acquisition unit, and an abnormal sound detection unit;

[0095] The audio data acquisition unit is used to acquire the sound data of the normal operation of the air conditioner outdoor unit to obtain the first sound data, and acquire the interference sound data in the environment where the air conditioner is detected to obtain the second sound data;

[0096] The time-frequency feature extraction unit is used to pre-process the first sound data and the second sound data to obtain time-frequency features, and then obtain third sound data and fourth sound data;

[0097] The Gaussian mixture model training unit is configured to train a Gaussian mixture model using the fourth sound data to obtain a fourth sound data distribution, and to superimpose the sound data sampled from the fourth sound data distribution onto the third sound data to obtain fifth sound data and a corresponding sampling step label;

[0098] The neural network model acquisition unit is used to acquire a neural network model, and the neural network model includes an audio noise reduction module and an anomaly detection module;

[0099] The neural network model training unit is configured to train the neural network model using the third sound data and the fifth sound data, first fixing the weights of the audio noise reduction module, and using the third sound data to train and update the parameters of the anomaly detection module, then fixing the weights of the anomaly detection module, and using the fifth sound data and sampling step labels to train the parameters of the audio noise reduction module;

[0100] The abnormal sound detection model obtaining unit is used to participate in the updating method according to the neural network model, repeating multiple times until the neural network model converges, to obtain a trained air conditioner outdoor unit anti-interference abnormal sound detection model;

[0101] The abnormal sound detection unit is used to obtain the operating sound data of the air conditioner to be tested, obtain sixth sound data through preprocessing, and use the trained air conditioner outdoor unit anti-interference abnormal sound detection model to perform abnormality detection on the sixth sound data.

[0102] Other features in this embodiment of the present invention are the same as those in Example 1, and therefore will not be described again here.

[0103] Example 3

[0104] Based on the same concept, the present invention also provides a schematic diagram of a physical structure, such as Figure 5As shown, the server may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the steps of the abnormal sound detection method for the air conditioner outdoor unit based on the generative model.

[0105] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0106] Example 4

[0107] Based on the same concept, the present invention also provides a non-transitory computer-readable storage medium, which stores a computer program. The computer program includes at least one segment of code, which can be executed by a main control device to control the main control device to implement the steps of the abnormal sound detection method of the air conditioner outdoor unit based on the generation model.

[0108] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in this application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive).

[0109] Those skilled in the art will appreciate that all or part of the process steps in the above-described method embodiments can be implemented by a computer program instructing the relevant hardware. The program can be stored in a computer-readable storage medium, and when executed, the program can include the process steps in the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

[0110] It is obvious to those skilled in the art that the present application is not limited to the details of the above-mentioned exemplary embodiments, and that the present application can be implemented in other specific forms without departing from the spirit or basic characteristics of the present application. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present application is defined by the appended claims rather than the above description, and it is intended that all changes that fall within the meaning and scope of the equivalent elements of the claims are included in the present application. Any accompanying figure marks in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "comprising" does not exclude other modules or steps, and the singular does not exclude the plural. Multiple modules or devices stated in the system claims can also be implemented by one module or device through software or hardware. Words such as first and second are used to indicate names and do not indicate any particular order.

Claims

1. A method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model, characterized in that: The steps include: Acquire sound data of normal operation of the air conditioner outdoor unit to obtain first sound data, and acquire interference sound data in the environment where the air conditioner is tested to obtain second sound data; Preprocessing the first sound data and the second sound data to obtain time-frequency features, thereby obtaining third sound data and fourth sound data; training a Gaussian mixture model using the fourth sound data to obtain a fourth sound data distribution, and superimposing sound data sampled from the fourth sound data distribution onto the third sound data to obtain fifth sound data and a corresponding sampling step label; Obtaining a neural network model, wherein the neural network model includes an audio noise reduction module and an anomaly detection module; Training the neural network model using the third sound data and the fifth sound data, first fixing the weights of the audio noise reduction module, using the third sound data to train and update the parameters of the anomaly detection module, then fixing the weights of the anomaly detection module, and using the fifth sound data and sampling step labels to train the parameters of the audio noise reduction module; According to the neural network model update method, the method is repeated multiple times until the neural network model converges, thereby obtaining a trained air conditioner outdoor unit anti-interference abnormal noise detection model; Acquire the operating sound data of the air conditioner to be tested, and obtain sixth sound data through preprocessing, and use the trained anti-interference abnormal sound detection model of the air conditioner outdoor unit to perform abnormality detection on the sixth sound data.

2. The method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model according to claim 1, characterized in that: The obtaining of interference sound data in the environment where the air conditioner is detected includes: The multi-channel sound transmission device collects the running sound of the transmission device, the movement sound of the cylinder, the running sound of the motor and the movement and impact sound of the mechanical structure.

3. The method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model according to claim 1, characterized in that: The preprocessing of the first sound data and the second sound data is time-frequency feature transformation, and the processing process includes: First, the sound data is framed and windowed, with some overlap between frames. Then, a short-time Fourier transform is performed on each frame of data to obtain the logarithmic time-frequency feature map.

4. The method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model according to claim 1, characterized in that: The Gaussian mixture model is trained using the fourth sound data to determine the distribution of the fourth sound data. include: Initialize the Gaussian mixture model parameters and use the EM algorithm to iteratively solve the fourth sound data distribution, its density function Expressed as: Where x is the sound data, M is the number of components of the Gaussian mixture model, μ i is the mean of the corresponding component, Σ i is the covariance matrix of the corresponding component, n is the dimension of sample x, i is the i-th Gaussian distribution function, and T is the matrix transpose.

5. The method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model according to claim 4, characterized in that: According to the third sound data having N samples, t interference noise samples are obtained by randomly sampling each sample in the third sound data from the Gaussian mixture density function in t steps, and are superimposed one by one with the third sound data to obtain the fifth sound data containing N×t sound samples and N×t corresponding sampling step labels S.

6. The method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model according to claim 5, characterized in that: The fifth sound data is obtained by combining the third sound data and the density function The noise samples obtained by sampling are superimposed and described as: in, is the i-th sample in the third sound data, for The fifth sound data sample is generated by superimposing the interference noise obtained by sampling in the tth step, and the weight α t =1-β t , for is the noise obtained by sampling, β is the manually set parameter, j is the jth step; The sampling step label S is normalized to (0, 1).

7. The method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model according to claim 1, characterized in that: The audio noise reduction module is a diffusion model, and the anomaly detection module is a deep convolutional autoencoder model.

8. The method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model according to claim 1, characterized in that: Training the neural network model includes: First, the third sound data is input into the network, the weight of the audio noise reduction module is fixed, and the anomaly detection module reconstructs the third sound data to update the network. Then, the weight of the anomaly detection module is fixed, and the audio noise reduction module sequentially estimates the interference noise superimposed on the fifth sound data and generates an estimated sampling step corresponding to the interference noise. The fifth sound data is then subtracted from the estimated interference noise and re-input into the audio noise reduction module. After all the fifth sound data are estimated, the fifth sound data obtained in the last step is input into the anomaly detection module. This process is repeated until the neural network model converges. The loss function of the neural network model is determined according to the following formula: Where ∈ is the interference noise obtained by actual sampling, ∈ θ is the interference noise estimated by the audio noise reduction module, t is the true value of the interference noise sampling step, is the interference noise sampling step prediction value estimated by the audio noise reduction module, x is the sound data after noise reduction, Reconstructed sound data for the anomaly detection module; θ and γ are both weight parameters of network loss.

9. The method for detecting abnormal sound of an air conditioner outdoor unit based on a generative model according to claim 1, characterized in that: The trained anti-interference and abnormal noise detection model for air conditioner outdoor units includes: The acquired air conditioner outdoor unit operating sound data is preprocessed to obtain time-frequency features, which are then input into the trained air conditioner outdoor unit anti-interference abnormal sound detection model; first, the estimated noise ∈ θ , and the estimated sampling step Then, the time-frequency feature is subtracted from the estimated noise and then re-input into the audio noise reduction module until the estimated sampling step output by the audio noise reduction module is If the value is less than the threshold th_denoise~(0,1), the corresponding audio noise reduction module input is subtracted from the estimated interference noise and then input into the abnormality detection module to obtain the reconstruction error e; when the reconstruction error e is greater than the threshold th, it is judged that the air conditioner operation sound is abnormal, otherwise it is normal.

10. An abnormal sound detection system for an air conditioner outdoor unit based on a generative model, characterized in that: include: An audio data acquisition unit is configured to acquire sound data of normal operation of the air conditioner outdoor unit to obtain first sound data, and acquire interference sound data in the environment in which the air conditioner is tested to obtain second sound data; a time-frequency feature extraction unit, configured to pre-process the first sound data and the second sound data to obtain time-frequency features, and then obtain third sound data and fourth sound data; a Gaussian mixture model training unit, configured to train a Gaussian mixture model using the fourth sound data to obtain a fourth sound data distribution, and to superimpose sound data sampled from the fourth sound data distribution onto the third sound data to obtain fifth sound data and a corresponding sampling step label; A neural network model acquisition unit, configured to acquire a neural network model, wherein the neural network model includes an audio noise reduction module and an anomaly detection module; a neural network model training unit, configured to train the neural network model using the third sound data and the fifth sound data, first fixing the weights of the audio noise reduction module, using the third sound data to train and update the parameters of the anomaly detection module, then fixing the weights of the anomaly detection module, and training the parameters of the audio noise reduction module using the fifth sound data and sampling step labels; an abnormal sound detection model obtaining unit, configured to obtain a trained anti-interference abnormal sound detection model for an air conditioner outdoor unit by repeatedly updating the neural network model until the neural network model converges; The abnormal sound detection unit is used to obtain the operating sound data of the air conditioner to be tested, obtain sixth sound data through preprocessing, and perform abnormality detection on the sixth sound data using the trained anti-interference abnormal sound detection model of the air conditioner outdoor unit.

Images