Intelligent fan abnormal sound discrimination and detection method based on time-frequency diagram and deep learning

Abstract

The application relates to a fan abnormal sound intelligent discrimination detection method and system based on a time-frequency graph and deep learning. The method comprises the following steps: collecting fan sound signals and fan rotating speed signals; identifying the fan sound signals and intercepting stationary section sound signals; performing A-weighting processing on the stationary section sound signals to obtain sound characteristics reflecting the loudness perception of human ears; performing short-time Fourier transform on the sound signals subjected to the A-weighting processing to obtain a time-frequency graph; inputting the time-frequency graph into a ResNet18 deep learning model to perform fan abnormal sound classification, so as to obtain an intelligent detection result of the fan running state. The application realizes intelligent detection of the fan running state.

Description

TECHNICAL FIELD

[0001] The application relates to a fan abnormal sound intelligent discrimination detection method based on a time-frequency graph and deep learning, and belongs to the fan detection field. BACKGROUND

[0002] In the production and quality control process of fans, fan abnormal sound detection is a key link to ensure product quality and performance. The traditional manual sound detection method has been widely used for a long time. Detection personnel judge whether the fan has abnormal sound by experience and hearing. However, this manual method has many limitations. On the one hand, manual detection is highly dependent on the professional skills, mental state and experience level of the detection personnel. The judgment standards of different detection personnel are difficult to keep completely consistent, which easily leads to subjective differences and misjudgment of the detection results. On the other hand, with the continuous expansion of fan production scale and the increasing requirement of production efficiency, the low efficiency of manual detection gradually becomes a bottleneck restricting the optimization of production process and the improvement of production capacity. It cannot meet the demand for accurate and efficient detection of fan abnormal sound in a large-scale and fast production environment, and has poor traceability. SUMMARY

[0003] The application provides a fan abnormal sound intelligent discrimination detection method based on a time-frequency graph and deep learning, which aims to at least solve one of the technical problems existing in the prior art.

[0004] The technical scheme of the application relates to a fan abnormal sound intelligent discrimination detection method based on a time-frequency graph and deep learning. According to the method of the application, the following steps are included:

[0005] S100, collecting fan sound signals and fan rotating speed signals;

[0006] S200, identifying the fan sound signals and intercepting the smooth section sound signals; performing A-weighting processing on the smooth section sound signals to obtain sound characteristics reflecting the loudness perception of human ears;

[0007] S300, performing short-time Fourier transform on the sound signals after A-weighting processing to obtain a time-frequency graph; inputting the time-frequency graph into a ResNet18 deep learning model to perform fan abnormal sound classification, so as to obtain an intelligent detection result of the fan running state.

[0008] Further, in the step S200,

[0009] The A-weighting processing attenuates low-frequency sound and enhances high-frequency sound, wherein the frequency response function H A (f) is represented as follows:

[0010]

[0011] In the formula, f represents frequency.

[0012] Further, in the step S300, the ResNet18 deep learning model comprises an input layer, an initial convolutional layer, a maximum pooling layer, a residual module, a flat pooling layer and a full connection layer connected in sequence.

[0013] The short-time Fourier transform STFT (t, f) is expressed as follows:

[0014]

[0015] In the formula, t represents time; x(t) represents the original signal; w(t-τ) represents a window function, where τ is the center time point of the window function, indicating that the signal is analyzed in the frequency spectrum at this time position; and f represents frequency.

[0016] Further, in the step S300, the ResNet18 deep learning model comprises an input layer, an initial convolutional layer, a maximum pooling layer, a residual module, a flat pooling layer and a full connection layer connected in sequence.

[0017] Further, the residual module network structure of the ResNet18 deep learning model comprises a first stage module, a second stage module, a third stage module and a fourth stage module: wherein,

[0018] The first stage module contains two residual blocks, and the output channel is set to 64;

[0019] The second stage module contains two residual blocks, and the output channel is set to 128;

[0020] The third stage module contains two residual blocks, and the output channel is set to 256;

[0021] The fourth stage module contains two residual blocks, and the output channel is set to 512.

[0022] Further, the residual block contains two convolutional layers, each of which uses a 3x3 convolutional kernel, and each of which is connected to a ReLU activation function and a batch normalization.

[0023] Further, the ResNet18 deep learning model uses a cross-entropy loss function to calculate the difference between the predicted value and the true value, and the cross-entropy loss function is expressed as follows:

[0024]

[0025] In the formula, y i represents the true class label, and p i represents the probability output by the model.

[0026] Further, the ResNet18 deep learning model uses an Adam optimizer to update the network weights, and the update rule of the Adam optimizer is set as follows:

[0027]

[0028] where t represents time, denotes the current gradient, m t and v t denote the first and second moment estimations of the gradient at time t, respectively, and denote the first and second moment estimations after bias correction, respectively; β1 and β2 are decay factors, and denote the exponential power of decay factor β1 at step t and the exponential power of decay factor β2 at step t, respectively; α is the learning rate; θ t denotes the model parameter at step t, θ t+1 denotes the model parameter updated at the next step; ∈ is a constant to prevent division by zero error.

[0029] The technical scheme of the present application also relates to a computer readable storage medium, which stores program instructions, and the program instructions are executed by a processor to implement the above method.

[0030] The technical scheme of the present application also relates to a fan abnormal sound intelligent discrimination and detection system based on a time-frequency graph and deep learning, which comprises a computer device containing the above computer readable storage medium.

[0031] The present application has the following advantages:

[0032] The present application is a fan abnormal sound intelligent detection method based on short-time Fourier transform and deep learning model, which overcomes the shortcomings of traditional artificial sound detection. The method performs short-time Fourier transform on the collected variable speed signal and generates a time-frequency graph, and uses a ResNet18 deep learning model to train and classify the time-frequency graph, thereby realizing intelligent detection of the fan running state. The present application uses data preprocessing and short-time Fourier transform (STFT) to extract the time-frequency features of the fan signal, which can significantly reduce the influence of background noise. Combined with the ResNet18 deep learning model, the method can effectively eliminate artificial dependence and reduce artificial errors, and is conducive to ensuring the stability and reliability of the detection results, improving the detection efficiency, and realizing data traceability and quality control enhancement. The method can record complete detection data including original sound signals, abnormal sound characteristics, prediction results, and time, and can trace the quality of specific fans or batches, locate the root cause of problems, and assist quality improvement. BRIEF DESCRIPTION OF DRAWINGS

[0033] Fig. 1 is a basic flowchart according to the method of the present application.

[0034] Fig. 2is a neural network structure diagram of a ResNet18 deep learning model according to the method of the application. DETAILED DESCRIPTION

[0035] The concept, specific structure and technical effects of the present application will be described clearly and completely in combination with the embodiments and the drawings, so as to fully understand the purpose, scheme and effect of the present application.

[0036] It should be noted that, unless otherwise specified, when a certain feature is referred to as being "fixed", "connected" to another feature, it can be directly fixed, connected to the other feature, or indirectly fixed, connected to the other feature. The singular forms "a", "said" and "the" used herein are also intended to include the plural forms, unless the context clearly indicates otherwise. In addition, unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art. The terms used in the specification herein are only for the purpose of describing specific embodiments and are not intended to limit the present application. The term "and / or" used herein includes any combination of one or more related listed items.

[0037] It should be understood that although the terms first, second, third, etc. can be used in the present disclosure to describe various elements, these elements should not be limited to these terms. These terms are only used to distinguish one type of element from another. For example, without departing from the scope of the present disclosure, a first element can also be referred to as a second element, and similarly, a second element can also be referred to as a first element. The use of any and all examples or exemplary language provided herein (e.g., "for example", "as", etc.) is intended only to better illustrate the embodiments of the present application, and unless otherwise required, does not impose a limitation on the scope of the present application.

[0038] Reference Figs. 1-2 In some embodiments, the fan abnormal sound intelligent discrimination detection method based on time-frequency graph and deep learning according to the present application at least includes the following steps:

[0039] S100, collecting fan sound signals and fan speed signals;

[0040] S200, identifying the fan sound signals and intercepting the smooth section sound signals; performing A-weighting processing on the smooth section sound signals to obtain sound characteristics reflecting the loudness perception of human ears;

[0041] S300, performing short-time Fourier transform on the sound signals after A-weighting processing to obtain a time-frequency graph; inputting the time-frequency graph into a ResNet18 deep learning model to classify fan abnormal sound, to obtain an intelligent detection result of the fan running state.

[0042] The fan abnormal sound intelligent detection method and system based on short-time Fourier transform and a deep learning model of the application, by performing short-time Fourier transform on the collected variable speed signal and generating a time-frequency graph, and using a ResNet18 deep learning model to train and classify the time-frequency graph, the intelligent detection of the fan running state is realized.

[0043] In some embodiments, in order to reduce the interference of environmental noise when collecting fan sound signals, the fan is placed in an existing soundproof box, and the sound signal during its operation is collected by a high-sensitivity microphone. The microphone is vertically installed perpendicular to the fan and located 5 cm above the center of the fan and points to the center of the fan. The system sampling frequency is set to 48 kHz to ensure sufficient capture of high-frequency components of the signal, and the analog signal is converted to a digital signal through a high-precision A / D conversion module to ensure the integrity and accuracy of the data. Further, the system also synchronously collects fan speed pulse signals, with 2 pulses recorded per revolution, to achieve accurate alignment of the sound signal and the fan mechanical state in subsequent analysis. After the system completes the fan sound signal collection, the collected data is transmitted to the data preprocessing unit for further processing.

[0044] In some embodiments, the application identifies and intercepts the sound signal segment of the fan in the stable running state from the collected fan sound signal. It should be noted that during the start and stop process of the fan, the sound signal will be affected by various unstable factors, such as start current impact, speed change, etc. The signal of the speed stable section adopted by the application can better reflect the sound characteristics of the fan during normal operation, which is conducive to the subsequent abnormal sound detection.

[0045] In some embodiments, the application performs A-weighting processing on the obtained stable section fan sound signal, so that the processed signal is more consistent with the subjective perception of human ears to sound loudness, which helps to extract sound characteristics with more practical significance and relevance. It should be noted that the A-weighting processing method adopted by the application can simulate the weighting of human ears to different frequency sound loudness perception. Specifically, based on the equal loudness curve, in acoustic measurement, the different frequency components of the sound signal are weighted and processed, so that the measurement result is more consistent with the subjective perception of human ears. Further, the A-weighting processing method of the application attenuates the low-frequency sound to a certain extent and appropriately enhances the high-frequency sound, so as to reflect the characteristics of human ears in loudness perception.

[0046] Specifically, the frequency response function H A (f) is represented as follows:

[0047]

[0048] In the formula, f represents frequency, and its unit is Hz.

[0049] In some embodiments, the present application performs fan abnormal sound feature extraction and screening on the sound signal after A-weighting processing. Specifically, in fan abnormal sound detection, feature extraction and screening play a crucial role, and the fan abnormal sound features of the present application mainly include 1 / 12 octave, 15 constant bandwidth energy, average spectrum, sound pressure level, loudness, etc.

[0050] In an application embodiment, for the extraction of 1 / 12 octave fan abnormal sound features, 1 / 12 octave is an analysis method that divides the frequency range into more fine intervals, and has important applications in the field of fan abnormal sound detection. 1 / 12 octave divides the audible frequency range (usually 20Hz-20kHz) into smaller frequency bands, compared with traditional octave analysis, it can reveal the characteristic changes of sound at different frequencies more meticulously. This is very important for detecting fan abnormal sound, because the abnormal sound of fan may occur in a specific narrow frequency range, and 1 / 12 octave can more accurately locate these abnormal frequency regions.

[0051] The center frequency f of 1 / 12 octave ci According to a specific geometric progression distribution, the formula is:

[0052] f ci = f0·2 i / 12

[0053] In the formula, f0 is the initial frequency, which is set to 20Hz in this project; i is the serial number of the octave.

[0054] The frequency range of each octave is determined by the upper and lower limit frequencies f li and f ui , and the specific formula is:

[0055] f li = f ci ·2 -1 / 24

[0056] f ui = f ci ·2 1 / 24

[0057] The upper and lower limit frequencies are calculated based on the center frequency with a scale of 2 ±1 / 24 , respectively, to ensure that the range width of each frequency band is symmetric on the logarithmic scale. The octave frequency bandwidth is:

[0058] BW i = f ui -f li

[0059] Energy is usually the integral (or cumulative in the case of discrete sampling) of signal power spectral density (PSD):

[0060]

[0061] If a discrete spectral representation is used (power spectrum P(fk) resulting from FFT computation):

[0062]

[0063] where P(fk) = |X(fk)| 2 is the power value of the spectrum; Δf = fs / N is the spectral resolution, and N is the number of points of the FFT.

[0064] In some embodiments, the present application performs a short-time Fourier transform (STFT) on the A-weighted processed sound signal to obtain a time-frequency graph. It can be understood that many actual sound signals have non-stationarity, i.e., the frequency components of the sound signal change over time, in order to effectively analyze these non-stationary sound signals, the present application uses a short-time Fourier transform (STFT) for time-frequency analysis, converts the sound signal from the time domain to the time-frequency domain, thereby obtaining the distribution information of the signal in time and frequency, and through the STFT transform method, the frequency characteristics of the signal at different time points can be obtained simultaneously, revealing the trend of the signal spectrum changing over time. Wherein, the present application performs a short-time Fourier transform on the A-weighted processed signal to obtain a time-frequency graph, and the picture size of the time-frequency graph is 224x224x3.

[0065] It should be noted that the short-time Fourier transform (STFT) used by the present application can be used to analyze non-stationary signals, convert the signal from the time domain to the time-frequency domain, and simultaneously obtain the frequency components of the signal at different time points, thereby showing the change of the signal spectrum over time. Wherein, the STFT transform of the present application performs window function weighting on the signal, divides the signal into multiple short time periods, and performs Fourier transform in each short time period, thereby analyzing the frequency spectrum.

[0066] Specifically, the signal is represented as x(t), and the window function is represented as w(t), then the short-time Fourier transform STFT(t,f) is represented as follows:

[0067]

[0068] where t represents time; x(t) represents the original signal; w(t-τ) represents the window function, where τ is the center time point of the window function, indicating that the signal is analyzed in the frequency spectrum at this time position; f represents frequency.

[0069] Further, in the frequency spectrum, the frequency resolution Δf = Fs / N = 5 Hz, where Fs represents the sampling frequency, the value of Fs can be 48 kHz, and N is the window length, i.e. the number of FFT points, and the overlap ratio can be set to 90%. It should be noted that the window function type used in the present application can be a Blackman-Harris window function.

[0070] In some embodiments, the present application realizes intelligent detection of the fan running state by constructing a ResNet18 deep learning model to train and classify the time-frequency graph. It can be understood that ResNet18, as a deep residual network, has strong feature extraction and classification capabilities, and the present application can realize effective analysis and classification of fan sound signals by constructing the neural network structure of the ResNet18 deep learning model. Specifically, ResNet18 solves the gradient vanishing and representation bottleneck problems in the training process of deep networks by introducing residual connections, which allow the network to more easily propagate gradients and information during training, thereby enabling the network to be deeper and more effective.

[0071] Specifically, the network structure of the ResNet18 of the present application is composed of multiple parts, including an initial convolutional layer, a residual block, and a fully connected layer, as shown in Fig. 2 The network architecture is as follows:

[0072] Input layer (image): receiving image input, where the size of the input time-frequency graph is 224x224x3, where 3 represents the RGB channel.

[0073] Initial convolutional layer (conv): the convolution kernel size is 7x7, the stride is 2, and the output channel number is 64. The first convolutional layer adopts the ReLU activation function and is normalized to batch normalization (Batch Normalization). The ReLU function is represented as follows:

[0074]

[0075] Where x is the input signal and f(x) is the output signal after activation.

[0076] Max-pooling layer (maxpool): the pooling kernel size is 3x33, and the stride is 2.

[0077] Residual module: each residual block contains two convolutional layers, each with a 3x3 convolutional kernel, and each convolutional layer is connected with a ReLU activation function and batch normalization. The skip connection directly connects the input and output through an identity mapping, which alleviates the problem of gradient vanishing.

[0078] Avgpool and FC (Fully Connected Layer): the input feature number of the FC layer is 512 and the output feature number is 1000

[0079] In an application embodiment, the residual module network structure of the present application is composed of four stage modules, wherein the first stage module (Stage 1) contains 2 residual blocks, and the output channel number is 64; the second stage module (Stage 2) contains 2 residual blocks, and the output channel number is 128; the third stage module (Stage 3) contains 2 residual blocks, and the output channel number is 256; and the fourth stage module (Stage 4) contains 2 residual blocks, and the output channel number is 512.

[0080] In an application embodiment, the ResNet18 deep learning model of the present application. Specifically, after short-time Fourier transform processing, the generated time-frequency graph is used as the input of the deep learning model ResNet18 for fan abnormal sound classification. The resolution of the time-frequency graph is set to 224x224 pixels, and the RGB three-channel format is adopted to meet the input requirements of ResNet18. The labels of the images are labeled according to the artificial listening results, wherein "normal fan" is defined as class 1, and "abnormal sound fan" is defined as class 0.

[0081] Referring to Fig. 2 , the ResNet18 deep learning model of the present application is based on the ResNet residual network, mainly composed of a residual module group of convolutional layers, the input of the residual module is directly added to the output of the module through identity mapping, and the deep residual network is composed of multiple residual module groups, for example, the model of the present application is composed of four stage modules, and the residual network of each stage model can fit the error of the previous classifier (see Fig. 2 the basic structure of the residual module), so as to improve the classification ability.

[0082] Specifically, the ResNet18 deep learning model of the present application is composed of 17 convolutional layers and 1 fully connected layer, wherein the 17 convolutional layers are composed of one independent convolution and four residual module groups, i.e. including an initial convolutional layer, a first stage module, a second stage module, a third stage module and a fourth stage module. Among them, the network is composed of multiple stage residual modules, each stage residual module uses identity mapping to combine the input and output information of the module, the last residual module is connected with the fully connected layer (FC), so as to reduce the fitting error, and finally the Softmax operation is performed to obtain the probability of the class. Referring to Fig. 2 the network structure of the ResNet18 deep learning model of the present application, the specific parameters are shown in Table I. Further, the network parameter settings of the ResNet18 deep learning model of the present application are shown in Table II.

[0083]

[0084] Table I

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092] Table II

[0093] In some embodiments, in order to ensure the training effect and generalization ability of the model, all images and their corresponding labels are arranged into a standard dataset format and are randomly shuffled before use. Subsequently, the dataset is divided into a training set and a test set according to a ratio of 80% and 20%. The training set is used for optimization of model parameters, and the classification performance is improved by minimizing the loss function; the test set is used to evaluate the generalization ability of the model, and the hyperparameters are adjusted and verified during the training process. Through this standardized processing, the scientificity of the data and the reliability of the experimental results are ensured.

[0094] Specifically, the loss function of the ResNet18 deep learning model of the present application is a cross-entropy loss function, which is used to calculate the difference between the predicted value and the true value, and its formula is:

[0095]

[0096] where y i is the true class label, p i is the probability output by the model.

[0097] Further, the present application uses an optimization algorithm to update the network weights. Specifically, the present application uses the Adam optimizer, which has the characteristics of adaptive learning rate, can accelerate convergence and prevent overfitting. The update rule of the Adam optimizer is set as follows:

[0098]

[0099] In the formula, t represents time, represents the current gradient, m t and vt They represent the estimates of the first-order moment and the second-order moment of the gradient at time t, and They represent the first-order moment after bias correction and the second-order moment after bias correction respectively; β1 and β2 are attenuation factors, and They represent the exponential power of the decay factor β1 at the tth step and the exponential power of the decay factor β2 at the tth step respectively; α is the learning rate; θ t represents the model parameters at step t, θ t+1 represents the model parameters after the next update; ∈ is a constant to prevent division by zero errors.

[0100] Furthermore, the learning rate can be set to 0.001. As you can understand, the learning rate determines the step size of each update of the model parameters. A smaller learning rate ensures a smoother convergence process, but may result in longer training times. Furthermore, the batch size can be set to 64. Setting this to 64 means that 64 samples will be used to calculate the gradient and update the network parameters during each optimization. A smaller batch size can lead to unstable training, but it is computationally more efficient. Furthermore, the number of training epochs can be set to 50.

[0101] After the model training is completed, the model performance is tested and evaluated. Specifically, during the model evaluation process, the test set is input into the trained model, and the classification accuracy, confusion matrix and loss function are used as the main evaluation indicators. Among them, the classification accuracy is used to measure the proportion of samples correctly classified by the model, and is an intuitive indicator for evaluating the overall performance of the model. The confusion matrix provides detailed information on the classification results, which includes true positive examples, false positive examples, true negative examples and false negative examples to reflect the classification ability of the model from multiple dimensions. The loss function is used to quantify the difference between the model prediction and the actual label. The smaller the value, the closer the model's prediction result is to the true value. The present invention can comprehensively evaluate the classification performance of the model and its optimization effect by comprehensively analyzing these three indicators. It can be understood that after obtaining the optimal training parameters, the deep learning model based on ResNet18 together with its network structure and weight parameters will be saved to ensure that the model can be directly loaded and used in subsequent detection processes, avoiding the waste of time and resources caused by repeated training.

[0102] In the actual detection task, the application needs to standardize and preprocess the data and convert the features for the newly collected fan sound signal. First, the sound signal is intercepted to retain the stable running stage signal, and then the signal is A-weighted to better meet the human auditory perception characteristics. Next, the short-time Fourier transform (STFT) is used to convert the signal into a time-frequency spectrogram, which is used as the input data form of the model. These time-frequency spectrograms need to go through the same processing steps as the training stage to ensure that the input data format is consistent with the input requirements of the ResNet18 model, thereby maximizing the detection accuracy and stability of the model. After completing the feature conversion, the generated time-frequency spectrogram is input into the trained ResNet18 deep learning model. The model extracts multi-level spatial features from the input image through its convolutional layers, and uses its pre-trained feature representation and optimized fully connected layers to complete the classification prediction. In the final decision layer of the model, the weight parameters learned during training are combined to predict the class of the input data. By analyzing the output probability distribution, the input sample can be quickly determined as a normal fan or an abnormal fan.

[0103] The fan abnormal sound intelligent discrimination detection method and system based on time-frequency spectrogram and deep learning of the embodiment of the application are verified on the blind side. Specifically, after the ResNet18 model is trained and saved, three independent blind tests are performed to comprehensively evaluate the classification ability and generalization performance of the model in actual application. The blind test data is completely independent of the training set and the test set, strictly simulating the real detection scene to ensure the scientificity and objectivity of the evaluation results.

[0104] Further, in each blind test, the newly collected fan sound data is first preprocessed to generate a time-frequency spectrogram with the same size as the training. Then, the time-frequency spectrogram is input into the trained ResNet18 deep learning model for classification prediction. The three blind test results are shown in Tables 1, 2 and 3. According to the table data, the constructed ResNet18 model not only has high classification accuracy, but also has good robustness and generalization ability, which can effectively cope with complex scenes in actual industrial detection. The successful verification of the blind test results further proves the scientificity and rationality of the feature selection and parameter optimization strategies adopted in the model design, and provides a reference for the practical application of the model in industrial fan abnormal sound detection.

[0105]

[0106] Table 1

[0107]

[0108] It should be appreciated that the steps of the methods in accordance with the embodiments of the application can be implemented or performed by a computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer readable medium. The methods can use standard programming techniques. Each program can implement the functions described in the application using high-level procedural or object-oriented programming languages to communicate with a computer system. However, the programs can be implemented in assembly or machine language, if desired. In any case, the language can be a compiled or interpreted language. Moreover, the programs can be stored on or downloaded from computer program products, which can be tangible computer readable storage media.

[0109] Further, the operations of the processes described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described herein (or variations and / or combinations thereof) can be performed under the control of one or more computer systems configured with executable instructions (e.g., computer programs, one or more computer programs, or one or more applications) to perform the processes, by hardware, or combinations thereof. The computer programs include processor-executable instructions that are stored in memory.

[0110] Further, the methods can be implemented in any suitable type of computing platform operably connected to any suitable type of computing platform, including but not limited to a personal computer, a mini-computer, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, and the like. Aspects of the application can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated to the computing platform, such as a hard disk, an optical read and / or write storage medium, a RSM, a ROM, and the like, such that it can be read by a programmable computer to configure and operate the computer to perform the processes described herein when the storage medium or device is read by the computer. Further, the machine-readable code, or portions thereof, can be transmitted over wired or wireless networks. The application described herein includes these and other different types of non-transitory computer readable storage media when such media include instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. The application can also include the computer itself when programmed in accordance with the methods and techniques described in the application.

[0111] The computer programs can be applied to input data to perform the functions described herein, to transform the input data into output data stored to a non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In preferred embodiments of the application, the transformed data represents a physical and tangible object, including a particular visual depiction of a physical and tangible object produced on a display.

[0112] The above merely describes preferred embodiments of the present application, and the present application is not limited to the above-described embodiments. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application, as long as the same technical effects are achieved by the same means, shall be included in the protection scope of the present application. The technical solutions and / or embodiments within the protection scope of the present application can have various modifications and changes.

Claims

1. The intelligent detection method for abnormal fan noise is characterized by: The method comprises the following steps: S100, collecting fan sound signals and fan speed signals; S200, identifying the fan sound signal and intercepting the steady-state sound signal; performing A-weighting processing on the steady-state sound signal to obtain sound characteristics that reflect the loudness perception of the human ear; S300, performing a short-time Fourier transform on the sound signal after A-weighting processing to obtain a time-frequency graph; inputting the time-frequency graph into a ResNet18 deep learning model to perform fan abnormal noise classification to obtain a detection result of the fan operating status; In step S200: The A-weighting process attenuates low-frequency sounds and enhances high-frequency sounds, wherein the frequency response function H of the A-weighting is A (f) is expressed as follows: Where, f represents frequency; The ResNet18 deep learning model uses the Adam optimizer to update the network weights. The update rule of the Adam optimizer is set as follows: Where t represents time, Represents the current gradient, m t and v t They represent the estimates of the first-order moment and the second-order moment of the gradient at time t, and They represent the first-order moment after bias correction and the second-order moment after bias correction respectively; β1 and β2 are attenuation factors, and They represent the exponential power of the decay factor β1 at the tth step and the exponential power of the decay factor β2 at the tth step respectively; α is the learning rate; θ t represents the model parameters at step t, θ t+1 represents the model parameters after the next update; ∈ is a constant to prevent division by zero errors.

2. The method according to claim 1, characterized in that In step S300: The short-time Fourier transform STFT(t,f) is expressed as follows: Where t represents time; x(t) represents the original signal; w(t-τ) represents the window function, where τ is the center time point of the window function, indicating that the spectrum analysis of the signal is performed at this time position; f represents frequency.

3. The method according to claim 1, characterized in that In step S300, the ResNet18 deep learning model includes an input layer, an initial convolutional layer, a maximum pooling layer, a residual module, an average pooling layer and a fully connected layer connected in sequence.

4. The method according to claim 3, characterized in that The residual module network structure of the ResNet18 deep learning model includes a first-stage module, a second-stage module, a third-stage module, and a fourth-stage module: The first-stage module contains two residual blocks, and the output channels are set to 64; The second stage module contains two residual blocks, and the output channels are set to 128; The third stage module contains two residual blocks, and the output channel is set to 256; The fourth stage module contains two residual blocks, and the output channels are set to 512.

5. The method according to claim 4, characterized in that The residual block includes two convolutional layers, each of which uses a 3×3 convolution kernel, and each of which is connected to a ReLU activation function and batch normalization.

6. The method according to claim 1, characterized in that The ResNet18 deep learning model uses a cross-entropy loss function to calculate the difference between the predicted value and the true value. The cross-entropy loss function is expressed as follows: Where y i represents the true category label, p i Represents the probability of the model output. 7 . A computer-readable storage medium having program instructions stored thereon, wherein the program instructions are configured to implement the method according to claim 1 when executed by a processor.

8. Intelligent identification and detection system for abnormal fan noise, characterized by: include: A computer device comprising the computer-readable storage medium according to claim 7.

Images