A method for judging abnormal sound failure of industrial data analysis

Abstract

This invention belongs to the field of industrial data analysis technology, and specifically relates to a method for diagnosing abnormal noise faults in industrial data analysis. The method collects baseline fault sound source data and mixed sound source data containing interference from a multi-dimensional distributed sensor array to construct a multimodal sound source database. Interference in the mixed sound source data is separated and energy entropy values ​​are calculated. After data quality verification, the denoised data is compared with the baseline data to filter valid data. A fault location model is constructed to determine the spatial coordinates of the fault point, eliminating false location points. A dual-branch neural network is used to fuse spatial and sound source data to output the fault category and probability. Compared with existing technologies, this invention overcomes the limitations of considering only sound or location information, effectively removes interference, verifies data quality, achieves accurate fault location and category judgment, improves fault detection accuracy and maintenance efficiency, and ensures the safe and stable operation of industrial equipment.

Description

Technical Field

[0001] This invention belongs to the field of industrial data analysis technology, and in particular relates to a method for diagnosing abnormal noise faults in industrial data analysis. Background Technology

[0002] In the field of industrial data analytics, equipment fault detection has always been a crucial link in ensuring the stable operation of industrial production. Taking elevators as an example, existing technical solutions for diagnosing abnormal noises in elevators have limitations, often considering only the sound or location. Simply relying on sound analysis cannot accurately pinpoint the location of the fault, making it difficult to quickly and effectively handle and repair it. Other technologies focus on location positioning, using methods such as measuring the time difference of signals arriving at different sensors using sensor arrays to determine the fault location. However, this approach ignores the rich fault characteristic information carried by the sound data; relying solely on location information makes it difficult to accurately determine the specific type of fault. Summary of the Invention

[0003] This invention addresses the technical problems existing in the background art by proposing a method for diagnosing abnormal noise faults through industrial data analysis.

[0004] To achieve the above objectives, the technical solution adopted by the present invention includes the following steps:

[0005] S1. A multimodal sound source database is constructed by synchronously collecting baseline fault sound source data and mixed sound source data with interference under interference conditions through a multi-dimensional distributed sensor array.

[0006] S2. Perform interference separation on the mixed sound source data to generate denoised target fault sound source data, and calculate the energy entropy value of the denoised target fault sound source data.

[0007] S3. Compare the denoised target fault data with the baseline fault sound source data. If the residual index exceeds the threshold, discard this part of the data; otherwise, proceed to the fault location stage. The specific judgment process is as follows:

[0008] S31. Convert the current and subsequent target fault data and the baseline fault sound source data into Mel spectrum diagrams;

[0009] S32. Extract 256-dimensional deep feature vectors using a two-stream network;

[0010] S33. Calculate the weighted cosine similarity difference of the five-layer feature maps and generate... ;

[0011] S34, Set threshold ;

[0012] S35, when > Data deemed unusable for noise reduction is discarded.

[0013] S36, when The fault location process will then be carried out.

[0014] S4. Construct a fault location model and determine the spatial coordinates of the fault point;

[0015] S5. Combining the spatial coordinates of the fault point and the sound source data, output the fault category and probability.

[0016] Preferably, the form of interference in step S1 includes echo, noise, and human voice.

[0017] Preferably, the calculation method for interference separation of the mixed sound sources in step S2 is as follows: ,in This is the noise-reduced data for the fault sound source. For wavelet packet transform operators, The data consists of mixed sound sources, where A is the mixed sound source matrix. The independent sound source components to be solved are: This is the regularization term for the total variation.

[0018] Preferably, the energy entropy value of the target fault sound source data in step S2 is calculated as follows: Where N is the number of wavelet packet decomposition levels, satisfying , This represents the data energy of the k-th subband.

[0019] As a preferred option, a data quality verification stage is added after calculating the energy entropy value of the target fault sound source data and before comparing the target denoised data and the baseline fault sound source data to make a quality decision: When the decision is 1, proceed with the next comparison.

[0020] Where Q is the quality evaluation index, and the calculation method is as follows: ,in To normalize the energy entropy, For signal-to-noise ratio, Indicates the kurtosis index of the data;

[0021] Among them, threshold =0.7 , The average value from 10 tests is used as the baseline, and the gradient threshold is... .

[0022] Preferably, step S4 is implemented as follows:

[0023] S41. Construct a fault location model, the calculation formula is: ,in Used to calculate the distance difference between the fault point and the sensor. Let m be the frequency domain signal of the m-th sensor. The complex conjugate of the reference sensor signal;

[0024] S42. Calculate the time difference between the arrival of the fault sound source at each sensor based on the model;

[0025] S43. Based on the sensor's spatial coordinates, establish a hyperbolic variable equation system and solve it to obtain the coordinates of the fault point.

[0026] As a preferred method, after obtaining the coordinates of the fault point, it is necessary to judge the coordinate points and eliminate false positioning points that exceed the physical boundaries of the equipment; the calculation method for the judgment is as follows: ,in For the estimated fault point coordinates, For the geometric center of the equipment, The maximum allowable offset is 20% of the device feature size.

[0027] Preferably, the method for outputting the fault category probability in step S5 is to first input spatial coordinates and sound source data information, then use a dual-branch neural network to fuse the spatial and sound source data, and output the fault category and probability.

[0028] Compared with existing technologies, the advantages and positive effects of this invention are as follows: It utilizes a multi-dimensional distributed sensor array to collect interference-free and interference-containing sound source data to construct a multi-modal sound source database; it employs a unique interference separation algorithm to remove interference such as echoes, noises, and human voices, and calculates the energy entropy value of the target fault sound source data; it adds a data quality verification stage to comprehensively normalize energy entropy, signal-to-noise ratio, and data kurtosis indicators to make quality decisions to avoid the impact of low-quality data; it constructs a fault location model to calculate the spatial coordinates of the fault point and eliminates false location points; and it uses a dual-branch neural network to fuse spatial and sound source data to output the fault category and probability, thereby accurately locating the fault location, determining the fault category, improving the accuracy of fault detection and maintenance efficiency, and ensuring the safe and stable operation of equipment. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart illustrating the overall structure of the present invention. Detailed Implementation

[0031] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0032] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0033] In this embodiment, to accurately determine the location and type of abnormal noise faults in industrial equipment and overcome the limitations of traditional technologies that rely solely on sound or location information, this invention proposes a method for determining abnormal noise faults using industrial data analysis. Taking elevators as an example, during elevator operation, traditional fault diagnosis methods either rely solely on listening to the sounds emitted by the elevator, with maintenance personnel using their experience to identify sound characteristics to roughly determine the fault. However, this method struggles to pinpoint exactly which part of the elevator is affected—whether it's the mechanical structure inside the car or a device in the shaft. In large buildings where multiple elevators operate simultaneously, this limitation is even more pronounced. Alternatively, it may rely solely on position detection technology, such as using sensors to measure the time difference between signals arriving at different position sensors to determine the fault location. However, this ignores the rich fault clues contained in the sound. For instance, when the elevator traction machine malfunctions, the differences in sound characteristics produced by different fault types can reflect the severity and specific problem, but position detection technology cannot utilize this information, leading to inaccurate fault classification and impacting maintenance efficiency and elevator operational safety. The specific implementation process of this invention is as follows: Figure 1 As shown.

[0034] When applied to elevators, this invention first collects baseline fault sound source data under interference-free conditions (such as simulating normal operation), as well as mixed sound source data containing various interferences during actual operation (such as background noise from the machine room, echoes in the shaft, etc.), and constructs a multimodal sound source database. For example, when the elevator operates under various conditions such as no-load and full-load operation, and stops at different floors, the collected sound data is classified and stored in the database.

[0035] Furthermore, considering that technology often lacks effective interference separation methods when processing mixed sound source data, a common approach is simple filtering. However, this method can only remove interference with relatively obvious frequency characteristics, and cannot accurately remove complex and diverse interference such as echoes, background noise, and human voices. This invention performs interference separation on mixed sound source data, and the calculation formula is as follows: ,in This is the noise-reduced data for the fault sound source. For wavelet packet transform operators, The data consists of mixed sound sources, where A is the mixed sound source matrix. The independent sound source components to be solved are: This is the total variation regularization term. Meanwhile, the energy entropy value of the target fault sound source data is calculated as follows: Where N is the number of wavelet packet decomposition levels, satisfying , This represents the data energy of the k-th sub-band. An improved interference separation algorithm and the introduction of a scientific method for calculating energy entropy enable efficient removal of various interferences from mixed sound source data and comprehensive and accurate extraction of the characteristics of the target fault sound source data. This process effectively extracts the sound features truly related to elevator faults, eliminating the influence of external interference factors. For example, in a noisy machine room environment, the sound data after interference separation can clearly reveal the subtle abnormal friction sounds of the traction machine, which may be masked in the original mixed data, providing crucial information for accurate fault diagnosis.

[0036] Before proceeding to fault location and category determination, this invention adds a data quality verification stage. Quality decisions are then made. When the decision value is 1, proceed with the next comparison; where Q is the quality evaluation index, calculated as follows: ,in To normalize the energy entropy, For signal-to-noise ratio, This represents the kurtosis index of the data; where the threshold is... =0.7 , The average value from 10 tests is used as the baseline, and the gradient threshold is... This step ensures the high reliability of the data used for fault diagnosis, avoiding misjudgments due to data quality issues. For example, if data collected in a particular instance is subject to sudden and severe interference, and quality verification reveals that it does not meet requirements, then that data is discarded and not included in subsequent analysis, thereby guaranteeing the accuracy of the fault diagnosis results.

[0037] Next, the denoised target fault data and the baseline fault sound source data are compared. When the residual index exceeds a threshold, this data is discarded. The comparison method first converts the current target fault data and the baseline fault sound source data into Mel spectrograms. Mel spectrograms are a spectral representation method based on human auditory characteristics, which better reflects the energy distribution of sound signals at different frequencies and is closer to human perception of sound. Through this conversion, the originally complex sound data is transformed into a more easily analyzed and understood spectral form. Next, a 256-dimensional deep feature vector is extracted using a two-stream network. A two-stream network is a widely used model structure in deep learning, capable of simultaneously extracting features from different perspectives, improving the comprehensiveness and accuracy of feature extraction. Here, the two-stream network fully mines the deep features in the Mel spectrogram, condensing them into a 256-dimensional feature vector. This 256-dimensional deep feature vector contains rich information about the sound data, covering multiple aspects such as frequency, amplitude, and phase. Then, the weighted cosine similarity difference of the five-layer feature mapping is calculated, and a specific index is generated. In deep learning models, feature maps are the results obtained after the network extracts features from the input data at different layers. By calculating the weighted cosine similarity difference of the five-layer feature maps, the similarity between the denoised target fault data and the baseline fault sound source data can be compared more precisely. The weighted cosine similarity takes into account the differences in importance of different feature dimensions, making the calculation results more consistent with reality. To accurately determine the reliability of the data, a reasonable threshold also needs to be set. When the generated metrics Greater than the threshold If the data is found to be denoised and the target fault data is significantly different from the baseline fault source data, the reliability of the data is questionable, and it may have been affected by unknown interference or a problem occurred during the denoising process. In this case, to ensure the accuracy of subsequent fault assessments, the data needs to be discarded to prevent this unreliable data from negatively impacting the final result. When the indicator... Less than or equal to the threshold If the data is denoised, it indicates that the target fault data after noise reduction has a high degree of similarity to the baseline fault sound source data, and the data has high reliability, so it can proceed to the subsequent fault location process.

[0038] In the fault location process, the fault location model is first constructed, and the calculation formula is as follows: ,in Used to calculate the distance difference between the fault point and the sensor. Let m be the frequency domain signal of the m-th sensor. The complex conjugate of the reference sensor signal is used; then, the time difference between the arrival of the fault sound source at each sensor is calculated according to the model; based on the spatial coordinates of the sensors, a hyperbolic variable equation system is established. The sensor's spatial coordinates are c is the speed of light. To calculate the time difference between the sound source reaching each sensor, the coordinate points need to be assessed during the solution process to eliminate false positioning points that exceed the physical boundaries of the equipment. The calculation method for this assessment is as follows: ,in For the estimated fault point coordinates, For the geometric center of the equipment, To determine the maximum permissible offset, 20% of the device's characteristic dimension is used. Next, cluster analysis is performed on the solution set that satisfies the constraints, and the weighted centroid is taken as the final fault location.

[0039] Finally, the spatial coordinates of the fault location and the sound source data are used as input. The spatial coordinates reflect the location distribution of the fault within the equipment, while the sound source data contains the characteristics of the fault sound. Next, a dual-branch neural network begins operation, consisting of a spatial data branch and a sound source data branch. The spatial data branch receives the spatial coordinates and extracts features through a series of convolutional and pooling layers, mining spatial location features such as the relative position of the fault location to key components of the equipment. The sound source data branch processes the sound source data, similarly using convolution and pooling operations to obtain features such as the frequency and amplitude of the sound. Then, the features extracted from the two branches are fused. Generally, a concatenation method is used, connecting the feature vectors from different branches sequentially to form a new feature vector containing both spatial and sound source information. Finally, the fused feature vector is input into a fully connected layer for classification and probability calculation. The fully connected layer further integrates the features through operations on the weight matrix and bias terms, ultimately outputting the fault category and the probability corresponding to each category. This achieves accurate fault category judgment and probability prediction, providing maintenance personnel with a clear direction for maintenance and a reference for the probability of the fault. Applied to the elevator industry, it enables precise location and classification of abnormal elevator noise faults, significantly improving the accuracy of fault detection and maintenance efficiency, and effectively ensuring the safe and stable operation of elevators.

[0040] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method of industrial data analysis abnormal sound failure determination, characterized by, The method comprises the following steps: S1, synchronously collecting reference fault sound source data under non-interference condition and mixed sound source data containing interference by a multi-dimensional distributed sensor array, and constructing a multi-modal sound source database; S2, separating the interference from the mixed sound source data, generating target fault sound source data after noise reduction, and calculating the energy entropy value of the target fault sound source data after noise reduction; S3, comparing the target fault sound source data after noise reduction with the reference fault sound source data, discarding the data when the residual index exceeds the threshold, otherwise entering the fault positioning stage, and the specific judgment process is as follows: S31, converting the target fault sound source data after noise reduction and the reference fault sound source data into Mel spectrograms; S32, extracting a 256-dimensional deep feature vector by a double-flow network; S33, calculate the weighted cosine similarity difference of the five-layer feature mapping, generate ; S34, set threshold ; S35、when , the data is determined to be denoising failure data, and an operation of discarding the data is performed;​ S36、when a fault location procedure will be performed; S4, constructing a fault positioning model to determine the spatial coordinates of the fault point; S5, combining the spatial coordinates of the fault point and the sound source data information to output the fault category and probability.

2. The method of claim 1, wherein, The form of the interference in step S1 includes echo, noise and human voice.

3. The method of claim 1, wherein the method further comprises: The calculation manner of interference separation of the mixed sound source in the step S2 is as follows: Wherein is the de-noised fault sound source data, is a wavelet packet transform operator, is the mixed sound source data, and A is a mixed sound source matrix, is an independent sound source component to be solved, is a total variation regular term.

4. The method of claim 1, wherein the method further comprises: The energy entropy value of the target fault sound source data after denoising in the step S2 is calculated in the following manner: wherein N is the number of wavelet packet decomposition, satisfying , represents the data energy of the kth decomposition subband.

5. The method of claim 1, wherein the method further comprises: A data quality verification stage is added before calculating the energy entropy value of the de-noised target fault sound source data and before comparing the de-noised target fault sound source data with the benchmark fault sound source data, to make a quality decision: When the decision is 1, the following comparison is performed. Wherein, Q is a quality evaluation index, and the calculation method is: Wherein is a normalized energy entropy, is a signal-to-noise ratio, indicates a data kurtosis index; wherein the threshold value = 0.7 , is the average value over 10 tests as a reference value, the gradient threshold .

6. The method of claim 1, wherein the method further comprises: After the coordinates of the fault point are obtained, the coordinate point needs to be judged to eliminate the false positioning point beyond the physical boundary of the device; the calculation method of the judgment is: wherein is the estimated fault point coordinate, is the geometric center of the device, is the maximum allowed offset, which is 20% of the characteristic size of the device.

7. The method of claim 1, wherein the method further comprises: The implementation manner of outputting the fault category probability in step S5 is that the spatial coordinates and the sound source data information are input first, the spatial data and the sound source data are fused by a double-branch neural network, and the fault category and the probability are output.

Images