An AI-assisted based CPB threshold setting method and system

Abstract

The application belongs to the technical field of equipment fault monitoring, and relates to an AI-assisted CPB threshold setting method and system. The method comprises the following steps: obtaining first sensor monitoring data of a sample product, and constructing a label matrix for identifying the running state of the product; calculating the energy value of the first sensor monitoring data in each CPB frequency band thereof to obtain an energy matrix; constructing a CPB frequency band energy feature vector based on the energy matrix and the label matrix; obtaining second sensor monitoring data of a product to be tested, and calculating the energy distribution of the second sensor monitoring data in each CPB frequency band thereof to obtain a feature vector corresponding to the second sensor monitoring data; and comparing the feature vector corresponding to the second sensor monitoring data with the constructed CPB frequency band energy feature vector to determine the current running state category of the product to be tested. The application solves the problems of strong dependence on artificial experience and difficulty in selecting a feature frequency band in traditional CPB threshold setting.

Description

Technical Field

[0001] This invention belongs to the technical field of equipment fault monitoring, and more specifically, relates to an AI-assisted CPB threshold setting method and system. Background Technology

[0002] In industrial production settings, monitoring the vibration and noise of products during operation is crucial for preventing malfunctioning products from entering the market and securing market share. To improve the quality of delivered products, companies introduce advanced technologies for quality inspection during product operation. Common methods include human hearing, time-domain signal energy detection, frequency-domain CPB analysis, and AI algorithms based on time-frequency features. In engineering, human hearing, time-domain signal energy detection, and frequency-domain CPB analysis are common methods. In research, AI algorithms based on time-frequency features are gaining popularity, with universities, research institutes, and companies proposing solutions from different perspectives.

[0003] For example, Chinese patent document CN112879278A discloses a method for diagnosing pump station unit faults based on A-weighted noise signal analysis. This method monitors the noise sound pressure signal of the pump station unit, performs spectral analysis on the sound pressure signal, corrects the sound pressure level of each frequency component using an A-weighted network, and obtains the A-weighted noise sound pressure level by superimposing the energy of the sound pressure levels. When the A-weighted noise sound pressure level exceeds an alarm value, a 1 / 3 octave band spectral analysis is performed on the sound pressure signal to extract the energy features of each octave band. A deep extreme learning machine is then used to quickly and effectively learn these features and extract the implicit fault information of each feature, thereby achieving intelligent diagnosis of pump unit faults. This invention uses a method and approach based on extracting frequency band energy features using a 1 / 3 octave band CPB.

[0004] Chinese patent document CN115753984A discloses a method for extracting abnormal features of idler rollers based on acoustic signature spectrum separation. After preprocessing the detection signal, the sound signal is pre-emphasized, framed, and windowed in the time domain. Short-time average energy and amplitude, short-time zero-crossing rate, peak-to-peak value, and kurtosis features are extracted and combined to form time-domain feature 1. When the value of time-domain feature 1 exceeds a preset threshold, an idler roller abnormality warning 1 is issued. The preprocessed signal is then subjected to Fourier transform to extract energy, subband energy ratio, formant features and sharpness, and 1 / 3 octave band features in the frequency domain, which are combined to form frequency-domain feature 2. When the value of frequency-domain feature 2 exceeds a preset threshold, an idler roller abnormality warning 2 is issued. When both warnings 1 and 2 occur, an acoustic signature spectrum is generated from the Fourier-transformed signal. The acoustic signature spectrum is then separated using HPSS to obtain harmonic components and shock wave components, and MFCC transform is performed to obtain acoustic signature feature 3. When the value of acoustic signature feature 3 exceeds a preset threshold, an idler roller abnormality is confirmed, and an alarm is issued. This invention proposes the idea of ​​using 1 / 3 octave band features to achieve combined frequency-domain features.

[0005] Chinese patent document CN115468643A discloses an elevator environmental vibration and noise comprehensive testing instrument and its testing method. The comprehensive testing instrument detects the vibration and noise of the elevator during operation, calculates the 1 / 3 octave band vertical vibration acceleration level, equivalent sound level noise value, and 1 / 1 octave band equivalent sound level value. The sound pressure level is used to determine whether the elevator is operating normally. The sound pressure level calculation is based on the energy distribution of the CPB frequency band.

[0006] Chinese patent document CN117191384A discloses a method and device for gear running-in quality assessment and abnormal noise location based on acoustic-vibration coupling. It acquires vibration acceleration and noise signals during the running-in process of a radar transmission system, calculates the 1 / 3 octave band spectrum of gear structural noise and air noise, and the vibration acceleration level and noise sound pressure level. The gear running-in status is determined by calculating the Mahalanobis distance of the characteristic samples of the vibration acceleration level and noise level. If the gear running-in is poor, the frequency band with the smallest difference between the vibration 1 / 3 octave band and the noise 1 / 3 octave band is selected, and the signal is bandpass filtered and envelope demodulated. The gear abnormal noise fault identification result is obtained through a preset expert experience value.

[0007] In the field of sound and vibration signal analysis, CPB (Constant Percent Bandwidth) octave band analysis is one of the most common methods, typically divided into different octaves such as 1 / 1, 1 / 3, 1 / 6, 1 / 12, and 1 / 24. When analyzing sound and vibration signals, it's often unnecessary to analyze every single frequency. For convenience, the entire frequency band is divided into multiple segments for analysis, each corresponding to a center frequency, upper limit frequency, and lower limit frequency. The division of frequency band segments is not arbitrary but follows certain rules. When a product operates abnormally or has potential abnormalities, the signal characteristics monitored by sensors are often reflected in the frequency spectrum. By establishing a CPB frequency band characteristic template for qualified products, it can be used to identify and detect products with operational abnormalities.

[0008] Traditional CPB band feature-based product malfunction handling procedures include: Figure 1As shown. The CPB feature analysis module calculates the Fourier transform of the sensor signal and calls the algorithm to complete the CPB feature analysis. The CPB feature analysis can be performed using different octave bands such as 1 / 1, 1 / 3, 1 / 6, 1 / 12, and 1 / 24, depending on project needs. In most cases, 1 / 3 octave band is used. The CPB frequency band feature template is created manually. The user inputs the upper and lower limits of the energy value on the specified frequency band based on experience or laboratory test data. When the energy value of the signal collected by the sensor on the manually specified frequency band exceeds the set threshold range, the system calibrates it as abnormal and classifies the fault according to the manually selected specified frequency band. For washing machine motor products, when faults such as grinding noise, vibration, and shaft asymmetry occur, the typical frequency band characteristic is a center frequency of 160Hz with different energy value distributions.

[0009] Traditional methods for identifying product malfunctions based on CPB band features require a high level of human experience and necessitate repeated testing of CPB band features for products with different fault categories during experiments. This involves calculating common features from a large number of products with the same fault category and finally identifying the most representative band features. Finding the appropriate frequency band with fault characteristics manually through experimentation is difficult, as is setting threshold values ​​for that band. Due to the tedious workload and operational difficulties, manual methods often rely on the most representative bands, typically 1-3 bands, as the basis for judgment. This filters out features of abnormal products in other frequency bands, increasing the false negative rate when the manually specified band features are not very obvious.

[0010] Therefore, there is an urgent need to design a more intelligent method for setting the CPB threshold, which can be applied to the edge devices of the production equipment health monitoring system to solve the problems of the traditional CPB threshold setting method being highly dependent on human experience and difficult to select characteristic frequency bands. Summary of the Invention

[0011] The present invention aims to overcome at least one of the defects of the prior art and provide an AI-assisted CPB threshold setting method, which can be applied to the quality inspection of products before they leave the production line in manufacturing enterprises. It solves the problems of traditional CPB threshold setting being highly dependent on human experience and difficult to select characteristic frequency bands.

[0012] The present invention also provides an AI-assisted CPB threshold setting system.

[0013] The detailed technical solution of this invention is as follows: A method for setting a CPB threshold value based on AI assistance, the method comprising: S1. Obtain first data: Obtain the first sensor monitoring data of the sample product within a sampling period, and record the label of the first sensor monitoring data to construct a label matrix L for identifying the operating status category of the sample product; S2. Data processing: Perform Fourier transform on the monitoring data of the first sensor to obtain the CPB spectrum of the monitoring data of the first sensor, divide the CPB spectrum of the monitoring data of the first sensor into several CPB frequency bands, and calculate the energy value of the monitoring data of the first sensor in each CPB frequency band to obtain the energy matrix E1. S3. Construct CPB band energy feature vectors: Filter out the row vectors corresponding to the sample products in the tag matrix L under each operating state category from the energy matrix E1 to form a subset of the CPB energy matrix corresponding to the sample products under each operating state category. Process the column vectors of the CPB energy matrix subsets and construct CPB band energy feature vectors according to the energy distribution of each CPB band. The element values ​​of the feature vectors are energy distribution density values ​​or correlation values ​​with the main data distribution interval. S4. Obtain the second data and its feature vector: Obtain the second sensor monitoring data of the product under test in the current operating state, perform Fourier transform on the second sensor monitoring data to obtain the CPB spectrum of the second sensor monitoring data, divide the CPB spectrum of the second sensor monitoring data into several CPB frequency bands, and calculate the energy distribution of the second sensor monitoring data in each CPB frequency band to obtain the feature vector corresponding to the second sensor monitoring data. S5. CPB band feature comparison: Compare the feature vector corresponding to the monitoring data of the second sensor with the CPB band energy feature vector constructed in step S3 to determine the current operating status category of the product under test.

[0014] According to a preferred embodiment of the present invention, in step S2, a Fourier transform is performed on the first sensor monitoring data to obtain the CPB spectrum of the first sensor monitoring data as follows: (1); In equation (1), The CPB spectral lines representing the data monitored by the first sensor reflect the energy at the corresponding frequency points. Indicates the position of the spectral line, and , This indicates the number of samples of data monitored by the first sensor. Indicates the first One data sample, It represents the imaginary part of a complex number.

[0015] According to a preferred embodiment of the present invention, in step S2, the energy value of the first sensor monitoring data in each CPB frequency band is calculated as follows: (2); In equation (2), This indicates that the data monitored by the first sensor is at the [number]th [time]. Energy values ​​in each CPB band Indicates the first The lower limit frequency of each CPB band Indicates the first The upper limit frequency of each CPB band.

[0016] According to a preferred embodiment of the present invention, in step S3, the column vectors of the CPB energy matrix subset are processed to construct a CPB frequency band energy feature vector based on the energy distribution of each CPB frequency band, specifically including: Extract the column vector corresponding to each CPB frequency band from the subset of the CPB energy matrix; Obtain the maximum value of the column vector corresponding to each CPB band. and minimum value and based on the maximum value and minimum value Divide the column vector corresponding to each CPB frequency band into A data interval, denoted as ,in , And the interval between each data range is: (3); Calculate the distribution density of the first sensor monitoring data in each data interval, for the first... Data range Distribution density of data monitored by the first sensor for: (4); In equation (4), Indicates the first The number of samples of the first sensor monitoring data in each data interval This indicates the number of samples of data monitored by the first sensor; The energy distribution density of each CPB band is calculated according to formula (4). ,in, This indicates that the data monitored by the first sensor is at the [number]th [time]. The first CPB band Data range The distribution density on is ; Select the maximum value of the CPB band energy distribution density corresponding to each CPB band. As the feature vector element values ​​of the corresponding CPB frequency band, the energy feature vector of the CPB frequency band under each operating state category is obtained, where... .

[0017] According to a preferred embodiment of the present invention, in step S3, the column vectors of the CPB energy matrix subset are processed to construct a CPB frequency band energy feature vector based on the energy distribution of each CPB frequency band, specifically further including: Calculate the mean of the CPB band energy distribution density for each CPB band. and variance The calculation formula is: (5); (6); (7); Based on the mean of the CPB band energy distribution density corresponding to each CPB band and variance For each CPB band, the data is divided into sub-intervals, and (-∞, mean -3) are generated sequentially. variance X%], [mean - 3] variance X%, mean -2 variance X%], [mean-2] variance X%, mean -1 variance X%], [mean-1] variance X%, mean +1 variance X%], [mean + 1] variance X%, mean + 2 variance X%], [mean + 2] variance X%, mean +3 variance X%], [mean + 3] variance X%, +∞), X% corresponds to the artificial adjustment coefficient, and the positional relationship between the sub-intervals of energy statistics and the mean reflects the correlation between the data; Define the adjustment coefficient for each sub-interval, obtain the correlation value of the energy statistical interval held by each CPB band energy value, use it as the feature vector element, and construct the CPB band energy feature vector under each operating state category.

[0018] According to a preferred embodiment of the present invention, step S3 further includes integrating the CPB band energy feature vectors constructed under each operating state category into a CPB band feature template library, wherein each feature vector in the CPB band feature template library reflects an operating state category of a product.

[0019] According to a preferred embodiment of the present invention, in step S4, the energy distribution of the second sensor monitoring data in each CPB frequency band is as follows: ; Based on the energy distribution of the second sensor monitoring data across its respective CPB frequency bands, the characteristic values ​​of the second sensor monitoring data in each CPB frequency band are obtained as follows: (8); In equation (8), This indicates that the data monitored by the second sensor is in the first... Energy values ​​in each CPB band This indicates that the data monitored by the second sensor is in the first... Characteristic values ​​on each CPB band; The feature vector composed of the feature values ​​in each CPB band is used as the feature vector of the second sensor monitoring data: (9); In equation (9), This represents the feature vector corresponding to the monitoring data from the second sensor.

[0020] According to a preferred embodiment of the present invention, in step S5, the similarity distance between the feature vector corresponding to the second sensor monitoring data and the CPB band energy feature vector in the CPB band feature template library is calculated, and the current operating status category of the product under test is determined based on the minimum similarity distance.

[0021] In another aspect of the present invention, an AI-assisted CPB threshold setting system is provided, the system comprising: The first data acquisition module is used to acquire the first sensor monitoring data of the sample product within a sampling period and record the labels of the first sensor monitoring data to construct a label matrix L for identifying the operating status category of the sample product. The data processing module is used to perform Fourier transform on the monitoring data of the first sensor to obtain the CPB spectrum of the monitoring data of the first sensor, divide the CPB spectrum of the monitoring data of the first sensor into several CPB frequency bands, and calculate the energy value of the monitoring data of the first sensor in each CPB frequency band to obtain the energy matrix E1. The CPB band energy feature vector construction module is used to filter out the row vectors corresponding to the sample products in the tag matrix L under each operating state category from the energy matrix E1, forming a subset of the CPB energy matrix corresponding to the sample products under each operating state category. The column vectors of the CPB energy matrix subset are processed, and a CPB band energy feature vector is constructed according to the energy distribution of each CPB band. The element values ​​of the feature vector are energy distribution density values ​​or correlation values ​​with the distribution interval of the main data. The second data acquisition module is used to acquire the second sensor monitoring data of the product under test in the current operating state, perform Fourier transform on the second sensor monitoring data to obtain the CPB spectrum of the second sensor monitoring data, divide the CPB spectrum of the second sensor monitoring data into several CPB frequency bands, and calculate the energy distribution of the second sensor monitoring data in each CPB frequency band to obtain the feature vector corresponding to the second sensor monitoring data. The CPB band feature comparison module is used to compare the feature vector corresponding to the monitoring data of the second sensor with the CPB band energy feature vector constructed in step S3 to determine the current operating status category of the product under test.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention provides an AI-assisted CPB threshold setting method and system, which constructs a feature vector template by mining the CPB frequency band energy distribution characteristics of sample products, thereby reducing the reliance on manual labor; at the same time, based on the constructed CPB frequency band feature vector template, the system automatically calculates the distance between vectors, predicts the health status of the product under operating conditions, and realizes anomaly classification. The method of the present invention avoids the use of complex machine learning algorithms, has low complexity, is simple to implement, and can be directly deployed to edge terminal devices, supporting cloud-edge-device collaborative management.

[0023] (2) This invention can be applied to quality inspection of products before they leave the production line in manufacturing enterprises. Based on the algorithm of this invention, it automatically generates product frequency band energy feature vectors for different fault types. The feature vectors contain multiple frequency bands. For the 1 / 3 octave band processing local time, the feature vector dimension constructed by the algorithm of this invention is 43. Since the algorithm of this invention automatically calculates the dimension and element values ​​of the energy matrix, it solves the problems of strong dependence on human experience and difficulty in selecting feature frequency bands in the traditional CPB threshold setting. Traditional human experience values ​​usually select a few fixed frequency bands and find the upper and lower limits of product energy distribution on the selected frequency bands through a large number of experiments during the sample training stage. Attached Figure Description

[0024] Figure 1 This is a flowchart of the traditional process for handling product malfunctions based on CPB band feature identification.

[0025] Figure 2 This is an execution flowchart of the AI-assisted CPB threshold setting method described in this invention.

[0026] Figure 3 This is a schematic diagram of the CPB band feature mining and analysis process in Embodiment 1 of the present invention.

[0027] Figure 4 This is a histogram of the energy distribution density characteristics of the CPB band in Embodiment 1 of the present invention. Detailed Implementation

[0028] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0029] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0030] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0031] Where there is no conflict, the embodiments and features described herein can be combined with each other.

[0032] Conducting operational quality testing before products leave the production line can prevent defective products from entering the market. Conventional testing methods involve collecting vibration or noise signals during product operation and identifying and classifying product anomalies by analyzing the characteristics of these signals. CPB analysis is a common processing method, especially for noise monitoring signals. Traditional CPB analysis requires manually setting detection thresholds and filtering frequency bands with obvious characteristics. This operation demands a high level of human experience and increases the complexity of manual operation.

[0033] For this application scenario, the present invention provides an AI-assisted CPB threshold setting method to solve the quality inspection problem before products leave the production line in manufacturing enterprises.

[0034] The core of this invention is a method for constructing CPB band energy feature vectors, which is constructed based on the energy distribution of each CPB band in the sensor monitoring data. The element values ​​of the feature vector are energy distribution density values ​​or correlation values ​​with the distribution interval of the main data.

[0035] In the present invention, firstly, based on the monitoring data of sample products collected by a sensor, CPB frequency band feature mining analysis is performed thereon to obtain an energy feature vector of the monitoring data of the sample products on each CPB frequency band, and through data accumulation and labeling, label data of the sample products under various operating conditions are obtained; then, a CPB frequency band energy feature vector is constructed by combining the energy feature vector of the monitoring data of the sample products on each CPB frequency band and the label data under various operating conditions.

[0036] Afterwards, based on the monitoring data of a product to be tested under the current operating condition collected by the sensor, CPB frequency band feature mining analysis is performed thereon to obtain an energy feature vector of the monitoring data of the product to be tested under the current operating condition on each CPB frequency band; then, by calculating the similarity distance between the energy feature vector of the monitoring data of the product to be tested under the current operating condition on each CPB frequency band and a CPB frequency band feature vector template, the current operating condition category of the product to be tested is further determined, realizing anomaly detection and classified management of product operating conditions, and further realizing identification and early warning of product health risks.

[0037] The overall architecture of the present invention has a self-learning and self-iteration function during actual operation, and can be started only by importing a small number of fault samples and clearly marking fault classifications in the cold start stage. The present invention is convenient in technical implementation, small in calculation amount and low in complexity, and is suitable for being applied to edge devices of a production equipment health monitoring system, which solves the problems that the traditional CPB threshold setting is highly dependent on manual experience and it is difficult to select characteristic frequency bands.

[0038] The AI-assisted CPB threshold setting method and system of the present invention will be further described below with reference to specific embodiments.

[0039] Embodiment 1 Reference Figure 2 , the present embodiment provides an AI-assisted CPB threshold setting method, comprising: S1, acquiring first data: acquiring first sensor monitoring data of a sample product within a sampling period, and recording labels of the first sensor monitoring data, so as to construct a label matrix L for identifying operating condition categories of the sample product.

[0040] In the embodiment, the labels are manually labeled, and for the initially constructed label matrix, a deduplication operation is further required for elements therein, and finally the label matrix L for identifying the operating condition categories of the sample product is obtained: .

[0041] Taking a certain motor device as an example herein, the operating condition categories of the sample product in the label matrix L can be {normal, jitter, grinding noise, crack, shaft asymmetry}.

[0042] S2. Data processing: Perform Fourier transform on the monitoring data of the first sensor to obtain the CPB spectrum of the monitoring data of the first sensor. Divide the CPB spectrum of the monitoring data of the first sensor into several CPB frequency bands and calculate the energy value of the monitoring data of the first sensor in each CPB frequency band to obtain the energy matrix E1.

[0043] Specifically, for the sensor monitoring data of each sample product, the first sensor monitoring data of all sample products obtained within one sampling period can be expressed as: , This represents the set of data monitored by all the first sensors. Indicates the first One data sample, This indicates the number of samples of the first sensor monitoring data collected within a sampling period.

[0044] Then, Fourier transform is performed on the collected data from the first sensor to obtain the corresponding CPB spectrum: (1); In equation (1), The CPB spectral lines representing the data monitored by the first sensor reflect the energy at the corresponding frequency points. Indicates the position of the spectral line, and , This indicates the number of samples of data monitored by the first sensor. Indicates the first One data sample, It represents the imaginary part of a complex number.

[0045] Next, the CPB spectrum of the first sensor monitoring data is divided into several CPB frequency bands. In this embodiment, the CPB frequency band division follows national standards, and Table 1 below shows the frequency band division results for CPB 1 / 3 octave bands.

[0046] Table 1: CPB 1 / 3 octave band division (unit: Hz)

[0047] Finally, the energy value of the first sensor monitoring data in each CPB band is calculated, i.e.: (2); In equation (2), This indicates that the data monitored by the first sensor is at the [number]th [time]. Energy values ​​in each CPB band Indicates the first The lower limit frequency of each CPB band Indicates the first the upper limit frequency of a CPB frequency band.

[0048] Through the above formula, the energy values of the first sensor monitoring data of all sample products in each CPB frequency band can be calculated, and finally the energy matrix E1 is generated as: , wherein, represents the total number of sample products, represents the number of CPB frequency bands, that is, it represents the th sample product at the th CPB frequency band. Moreover, the th row element value in the label matrix L corresponds to the th row element in the energy matrix E1, and represents the operating status category of the th sample product in each CPB frequency band.

[0049] S3, Constructing CPB frequency band energy feature vectors: filtering out the row vectors corresponding to sample products in the label matrix L under each operating status category from the energy matrix E1, forming CPB energy matrix subsets corresponding to sample products under each operating status category, processing the column vectors of the CPB energy matrix subsets, and constructing CPB frequency band energy feature vectors according to the energy distribution of each CPB frequency band, wherein the element values of the feature vectors are energy distribution density values or values related to the main data distribution interval.

[0050] Ref Figure 3 , CPB frequency band feature mining analysis can mine data features for the sample energy values of each frequency band according to the frequency bands set by CPB or manually set frequency bands. The data features herein depend on the mean, variance and data density of energy distribution on the frequency band.

[0051] The CPB frequency band energy distribution can be processed according to product categories (abnormal types, models), the processing method can divide subsets, and sample products in each subset have the same attribute, for example, all belong to grinding noise fault motor products.

[0052] Continuing to take a certain motor equipment as an example herein, the operating status categories in the label matrix L are {normal, jitter, grinding noise, crack, shaft asymmetry}.

[0053] Row vectors with the operating status category of {normal} are filtered out from the energy matrix E1 to form a CPB energy matrix subset under normal operating conditions , namely: , wherein, , , respectively represent the st, the rd, the Taiwan motor, Indicates the number of CPB bands. Indicates that there is The motor is operating normally.

[0054] Similarly, row vectors with the operating state category {jitter} are filtered out from the energy matrix E1 to form a subset of the CPB energy matrix for jitter operating states. ,Right now: ,in, , , They represent the first Taiwan, No. Taiwan, No. Taiwan motor, Indicates that there is The operating state of the motor satisfies {vibration}.

[0055] Similarly, row vectors with the operating state category {grinding noise} are filtered out from the energy matrix E1 to form a subset of the CPB energy matrix under the grinding noise operating state. ; Filter out the row vectors with the operating state category {crack} from the energy matrix E1 to form a subset of the CPB energy matrix for the crack operating state. ; Filter out the row vectors with the running state category {axis asymmetric} from the energy matrix E1 to form a subset of the CPB energy matrix for axis asymmetric running states. .

[0056] In this embodiment, the column vectors of the CPB energy matrix subset are processed to construct CPB frequency band energy feature vectors based on the energy distribution of each CPB frequency band. This can be achieved using the following two methods: Method 1: Extract the energy feature vector of each CPB band based on the energy distribution density function of each CPB band.

[0057] Specifically: First, we extract the column vector corresponding to each CPB frequency band in the CPB energy matrix subset for each operating state category of the sample product. Here, we take the CPB energy matrix subset under jitter operating state as an example. For example, a subset of the CPB energy matrix Rewritten as: For the first The extracted column vectors for each CPB band are: ,in, Indicates the first The motor is in the first Energy of the CPB band.

[0058] Then obtain the maximum value of the column vector corresponding to each CPB band. and minimum value and based on the maximum value and minimum value Divide the column vector corresponding to each CPB frequency band into The data interval can be divided into equal parts, or it can be divided according to the division rules under the CPB 1 / 3 octave band set by the national standard, denoted as: ,in , And the interval between each data range is: (3).

[0059] Then calculate the distribution density of the first sensor monitoring data in each data interval, for the first... Data range Distribution density of data monitored by the first sensor for: (4); In equation (4), Indicates the first The number of samples of the first sensor monitoring data in each data interval This indicates the number of samples of data monitored by the first sensor.

[0060] Then, according to formula (4), the CPB band energy distribution density corresponding to each CPB band is calculated as follows: ,in, This indicates that the data monitored by the first sensor is at the [number]th [time]. The first CPB band Data range The distribution density on is The energy distribution density characteristics of the CPB band were plotted as a curve, as shown below. Figure 4 The histograms shown; the height of each histogram reflects the probability value of the data distribution in that frequency band.

[0061] Finally, the maximum value of the CPB band energy distribution density corresponding to each CPB band is selected. As the feature vector element values ​​of the corresponding CPB frequency band, the energy feature vector of the CPB frequency band under each operating state category is obtained, where... .

[0062] The CPB band feature vector template for the sample product under jitter operation, constructed based on this method, is as follows: .

[0063] Similarly, we can obtain the CPB band feature vector templates for other operating state categories: Under normal operating conditions, ; Under the operating conditions of the grinding mill, ; Under cracked operating conditions, ; In asymmetric operation mode, .

[0064] One of the core ideas of this embodiment is to obtain CPB band energy distribution feature vectors under different operating states of the product through the CPB band energy distribution density characteristics. Each feature vector corresponds to a certain operating state. With the feature vectors, the distance between them can be calculated, enabling product anomaly detection and fault classification.

[0065] Method 2: Based on the mean value of the CPB band energy distribution density corresponding to each CPB band. and variance Extract the energy feature vector of the CPB band.

[0066] Continuing with the CPB energy matrix subset under jittery operating conditions For example, consider a subset of the CPB energy matrix under jittery operating conditions. Perform column processing operations.

[0067] Since the energy distribution in the CPB band is not uniform or normal, it is necessary to calculate the mean and variance of energy based on the energy distribution density function of each band.

[0068] For the The energy distribution density of the CPB band is as follows: .

[0069] Then, the mean of the data in this frequency band and variance They are respectively: (5); (6); (7); The statistical characteristics of the monitoring data are recorded as follows: .

[0070] Based on the mean of the CPB band energy distribution density corresponding to each CPB band By determining the feature vector elements of the corresponding CPB frequency band, we can obtain the CPB frequency band energy feature vector of the sample product under each operating state category.

[0071] The CPB band feature vector template for the sample product under jitter operation, constructed based on this method, is as follows: .

[0072] Similarly, we can obtain the CPB band feature vector templates for other operating state categories: Under normal operating conditions, ; Under the operating conditions of the grinding mill, ; Under cracked operating conditions, ; In asymmetric operation mode, .

[0073] Furthermore, based on the mean and variance of the energy distribution, the CPB band feature mining under each anomaly category sequentially generates energy statistical sub-intervals (-∞, mean -3). variance X%], [mean - 3] variance X%, mean -2 variance X%], [mean-2] variance X%, mean -1 variance X%], [mean-1] variance X%, mean +1 variance X%], [mean + 1] variance X%, mean + 2 variance X%], [mean + 2] variance X%, mean +3 variance X%], [mean + 3] variance X%, +∞). X% corresponds to the artificial adjustment coefficient. The positional relationship between the energy statistics sub-interval and the mean reflects the correlation between the data.

[0074] For example, for the first Data can be divided into intervals within a CPB band, for example, the data distribution can be divided into 7 intervals: , , , , , , Each interval reflects its distance from the mean; this distance can be described as correlation.

[0075] By defining the adjustment coefficient for each sub-interval, the relevant values ​​of the energy statistics interval (main data interval) held by the energy value of each CPB band can be obtained. These values ​​can then be used as feature vector elements to construct the CPB band energy feature vector under each operating state category.

[0076] The second core idea of ​​this embodiment is to use the mean value of the energy distribution characteristics of the CPB band. and variance Obtain CPB band energy feature vectors under different operating states of the product. Each feature vector corresponds to a specific operating state. With these feature vectors, the distances between them can be calculated, enabling product anomaly detection and fault classification.

[0077] It should be understood that for features mined based on density distribution functions, the element values ​​of the feature vector are the distribution density values ​​corresponding to the energy values ​​of that frequency band. For features mined based on mean and variance, the element values ​​of the feature vector are the correlation values ​​of the energy statistical interval held by the energy values ​​of that frequency band.

[0078] Finally, the CPB band energy feature vectors constructed under each operating state category can be integrated into a CPB band feature template library, where each feature vector in the CPB band feature template library reflects an operating state category of a product.

[0079] S4. Obtain the second data and its feature vector: Obtain the second sensor monitoring data of the product under test in the current operating state, perform Fourier transform on the second sensor monitoring data to obtain the CPB spectrum of the second sensor monitoring data, divide the CPB spectrum of the second sensor monitoring data into several CPB frequency bands, and calculate the energy distribution of the second sensor monitoring data in each CPB frequency band to obtain the feature vector corresponding to the second sensor monitoring data.

[0080] Similar to step S1, the monitoring data from the second sensor of the product under test in its current operating state is acquired, and then divided into CPB frequency bands after FFT transformation. Details will not be elaborated here.

[0081] After obtaining the monitoring data of the second sensor of the product under test in its current operating state, the feature vector extraction is initiated after FFT transformation and CPB feature analysis. The extraction process can be implemented using the two methods mentioned above.

[0082] Specifically, Method 1 is as follows: First, the energy distribution of the second sensor's monitoring data across its respective CPB frequency bands is calculated: ,here Indicates the first The energy value in each CPB band is calculated using the same formula as formula (2), that is: Based on this, the energy value of each CPB band can be calculated.

[0083] For the For each CPB frequency band, the CPB frequency band energy distribution density can be obtained using the CPB frequency band feature vector template: .

[0084] By combining the CPB band feature vector template of this frequency band, the monitoring data of the second sensor can be obtained at the [missing information]. The characteristic values ​​for each CPB band are: (8); In equation (8), This indicates that the data monitored by the second sensor is in the first... Energy values ​​in each CPB band This indicates that the data monitored by the second sensor is in the first... Characteristic values ​​on each CPB band.

[0085] Based on this, the characteristic values ​​of the second sensor monitoring data in each CPB frequency band can be obtained, and then the feature vector composed of the characteristic values ​​of the second sensor monitoring data in each CPB frequency band can be obtained as follows: (9); Right now, This represents the feature vector corresponding to the monitoring data from the second sensor.

[0086] Method two is: First, the energy distribution of the second sensor monitoring data across its various CPB frequency bands is obtained as follows: .

[0087] Then, based on the mean and variance of the energy distribution, the distribution is divided into 7 sub-intervals, and the correlation coefficient between the sub-intervals is defined as follows: Correlation is ; Correlation is ; Correlation is ; Correlation is ; Correlation is ; Correlation is ; Correlation is .

[0088] Then, the data monitored by the second sensor in the 1st... The characteristic values ​​for each CPB frequency band are: ; ; ; ; ; ; ; Finally, the feature vector composed of the feature values ​​of the second sensor monitoring data across all CPB frequency bands is obtained as follows: That is, the feature vector corresponding to the monitoring data of the second sensor.

[0089] S5. CPB band feature comparison: Compare the feature vector corresponding to the monitoring data of the second sensor with the CPB band energy feature vector constructed in step S3 to determine the current operating status category of the product under test.

[0090] CPB band feature comparison supports full-band comparison and interval band comparison modes. The comparison is based on the CPB threshold range or spatial vector distance.

[0091] In this embodiment, cosine similarity or Mahalanobis distance can be used to calculate the similarity distance between the feature vector corresponding to the monitoring data of the second sensor and the energy feature vector of the CPB band in the CPB band feature template library, and then the current operating status category of the product under test can be determined based on the minimum similarity distance.

[0092] Specifically, the CPB band feature vector templates for sample products under various operating status categories, stored in the CPB band feature vector template library, are as follows: ; ; ; ; .

[0093] Feature vector corresponding to the data monitored by the second sensor for: .

[0094] The distance between feature vectors can be calculated using cosine similarity or Mahalanobis distance. The feature vectors corresponding to the second sensor monitoring data... With template feature vector , , , , The distances are respectively , , , , Then sort these 5 distance values ​​and take the minimum distance. Assume... If the value is the smallest, it means that the characteristics of the second sensor monitoring data of the product under test in its current operating state are most similar to the characteristics of the grinding noise abnormality, and it can be determined that the product under test is currently experiencing a grinding noise abnormality.

[0095] It should be understood that when performing feature comparison as described above, it is necessary to poll all feature vectors in the template library and match the feature vector that is closest to the CPB band energy distribution characteristics of the sensor data of the product under test. The operating status category corresponding to this feature vector is the operating status category of the product under test.

[0096] Example 2 This embodiment provides an AI-assisted CPB threshold setting system, the system comprising: The first data acquisition module is used to acquire the first sensor monitoring data of the sample product within a sampling period and record the labels of the first sensor monitoring data to construct a label matrix L for identifying the operating status category of the sample product. The data processing module is used to perform Fourier transform on the monitoring data of the first sensor to obtain the CPB spectrum of the monitoring data of the first sensor, divide the CPB spectrum of the monitoring data of the first sensor into several CPB frequency bands, and calculate the energy value of the monitoring data of the first sensor in each CPB frequency band to obtain the energy matrix E1. The CPB band energy feature vector construction module is used to filter out the row vectors corresponding to the sample products in the tag matrix L under each operating state category from the energy matrix E1, forming a subset of the CPB energy matrix corresponding to the sample products under each operating state category. The column vectors of the CPB energy matrix subset are processed, and a CPB band energy feature vector is constructed according to the energy distribution of each CPB band. The element values ​​of the feature vector are energy distribution density values ​​or correlation values ​​with the distribution interval of the main data. The second data acquisition module is used to acquire the second sensor monitoring data of the product under test in the current operating state, perform Fourier transform on the second sensor monitoring data to obtain the CPB spectrum of the second sensor monitoring data, divide the CPB spectrum of the second sensor monitoring data into several CPB frequency bands, and calculate the energy distribution of the second sensor monitoring data in each CPB frequency band to obtain the feature vector corresponding to the second sensor monitoring data. The CPB band feature comparison module is used to compare the feature vector corresponding to the monitoring data of the second sensor with the CPB band energy feature vector constructed in step S3 to determine the current operating status category of the product under test.

[0097] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solutions of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A method for setting a CPB threshold value based on AI assistance, characterized in that, The method includes: S1. Obtain first data: Obtain the first sensor monitoring data of the sample product within a sampling period, and record the label of the first sensor monitoring data to construct a label matrix L for identifying the operating status category of the sample product; S2. Data processing: Perform Fourier transform on the monitoring data of the first sensor to obtain the CPB spectrum of the monitoring data of the first sensor, divide the CPB spectrum of the monitoring data of the first sensor into several CPB frequency bands, and calculate the energy value of the monitoring data of the first sensor in each CPB frequency band to obtain the energy matrix E1. S3. Construct CPB band energy feature vectors: Filter out the row vectors corresponding to the sample products in the tag matrix L under each operating state category from the energy matrix E1 to form a subset of the CPB energy matrix corresponding to the sample products under each operating state category. Process the column vectors of the CPB energy matrix subsets and construct CPB band energy feature vectors according to the energy distribution of each CPB band. The element values ​​of the feature vectors are energy distribution density values ​​or correlation values ​​with the main data distribution interval. S4. Obtain the second data and its feature vector: Obtain the second sensor monitoring data of the product under test in the current operating state, perform Fourier transform on the second sensor monitoring data to obtain the CPB spectrum of the second sensor monitoring data, divide the CPB spectrum of the second sensor monitoring data into several CPB frequency bands, and calculate the energy distribution of the second sensor monitoring data in each CPB frequency band to obtain the feature vector corresponding to the second sensor monitoring data. S5. CPB band feature comparison: Compare the feature vector corresponding to the monitoring data of the second sensor with the CPB band energy feature vector constructed in step S3 to determine the current operating status category of the product under test.

2. The AI-assisted CPB threshold setting method according to claim 1, characterized in that, In step S2, the Fourier transform of the first sensor monitoring data is performed to obtain the CPB spectrum of the first sensor monitoring data: （1）； In equation (1), The CPB spectral lines representing the data monitored by the first sensor reflect the energy at the corresponding frequency points. Indicates the position of the spectral line, and , This indicates the number of samples of data monitored by the first sensor. Indicates the first One data sample, It represents the imaginary part of a complex number.

3. The AI-assisted CPB threshold setting method according to claim 1, characterized in that, In step S2, the energy value of the first sensor monitoring data in each CPB frequency band is calculated: （2）； In equation (2), The CPB spectral lines representing the data monitored by the first sensor reflect the energy at the corresponding frequency points. Indicates the position of the spectral line, and , This indicates the number of samples of data monitored by the first sensor. This indicates that the data monitored by the first sensor is at the [number]th [time]. Energy values ​​in each CPB band Indicates the first The lower limit frequency of each CPB band Indicates the first The upper limit frequency of each CPB band.

4. The AI-assisted CPB threshold setting method according to claim 3, characterized in that, In step S3, the column vectors of the CPB energy matrix subset are processed to construct a CPB frequency band energy feature vector based on the energy distribution of each CPB frequency band. Specifically, this includes: Extract the column vector corresponding to each CPB frequency band from the subset of the CPB energy matrix; Obtain the maximum value of the column vector corresponding to each CPB band. and minimum value and based on the maximum value and minimum value Divide the column vector corresponding to each CPB frequency band into A data interval, denoted as ,in , And the interval between each data range is: （3）； Calculate the distribution density of the first sensor monitoring data in each data interval, for the first... Data range Distribution density of data monitored by the first sensor for: （4）； In equation (4), Indicates the first The number of samples of the first sensor monitoring data in each data interval This indicates the number of samples of data monitored by the first sensor; The energy distribution density of each CPB band is calculated according to formula (4). ,in, This indicates that the data monitored by the first sensor is at the [number]th [time]. The first CPB band Data range The distribution density on is ; Select the maximum value of the CPB band energy distribution density corresponding to each CPB band. As the feature vector element values ​​of the corresponding CPB frequency band, the energy feature vector of the CPB frequency band under each operating state category is obtained, where... .

5. The AI-assisted CPB threshold setting method according to claim 4, characterized in that, In step S3, the column vectors of the CPB energy matrix subset are processed to construct a CPB frequency band energy feature vector based on the energy distribution of each CPB frequency band. Specifically, this also includes: Calculate the mean value of the CPB band energy distribution density for each CPB band. and variance The calculation formula is: （5）； （6）； （7）； Based on the mean of the CPB band energy distribution density corresponding to each CPB band and variance For each CPB band, the data is divided into sub-intervals, and (-∞, mean -3) are generated sequentially. variance X%], [mean - 3] variance X%, mean -2 variance X%], [mean-2] variance X%, mean -1 variance X%], [mean-1] variance X%, mean +1 variance X%], [mean + 1] variance X%, mean + 2 variance X%], [mean + 2] variance X%, mean +3 variance X%], [mean + 3] variance X%, +∞), X% corresponds to the artificial adjustment coefficient, and the positional relationship between the sub-intervals of energy statistics and the mean reflects the correlation between the data; Define the adjustment coefficient for each sub-interval, obtain the correlation value of the energy statistical interval held by each CPB band energy value, use it as the feature vector element, and construct the CPB band energy feature vector under each operating state category.

6. The AI-assisted CPB threshold setting method according to claim 1, characterized in that, Step S3 further includes integrating the CPB band energy feature vectors under each constructed operating state category into a CPB band feature template library, wherein each feature vector in the CPB band feature template library reflects an operating state category of a product.

7. The AI-assisted CPB threshold setting method according to claim 4, characterized in that, In step S4, the energy distribution of the second sensor monitoring data across its respective CPB frequency bands is as follows: , Indicates the number of CPB bands; Based on the energy distribution of the second sensor monitoring data across its respective CPB frequency bands, the characteristic values ​​of the second sensor monitoring data in each CPB frequency band are obtained as follows: （8）； In equation (8), This indicates that the data monitored by the second sensor is in the first... Energy values ​​in each CPB band This indicates that the data monitored by the second sensor is in the first... Characteristic values ​​on each CPB band; The feature vector composed of the feature values ​​in each CPB band is used as the feature vector of the second sensor monitoring data: （9）； In equation (9), This represents the feature vector corresponding to the monitoring data from the second sensor.

8. The AI-assisted CPB threshold setting method according to claim 3, characterized in that, In step S5, the similarity distance between the feature vector corresponding to the second sensor monitoring data and the CPB band energy feature vector in the CPB band feature template library is calculated, and the current operating status category of the product under test is determined based on the minimum similarity distance.

9. An AI-assisted CPB threshold setting system, characterized in that, The system includes: The first data acquisition module is used to acquire the first sensor monitoring data of the sample product within a sampling period and record the labels of the first sensor monitoring data to construct a label matrix L for identifying the operating status category of the sample product. The data processing module is used to perform Fourier transform on the monitoring data of the first sensor to obtain the CPB spectrum of the monitoring data of the first sensor, divide the CPB spectrum of the monitoring data of the first sensor into several CPB frequency bands, and calculate the energy value of the monitoring data of the first sensor in each CPB frequency band to obtain the energy matrix E1. The CPB band energy feature vector construction module is used to filter out the row vectors corresponding to the sample products in the tag matrix L under each operating state category from the energy matrix E1, forming a subset of the CPB energy matrix corresponding to the sample products under each operating state category. The column vectors of the CPB energy matrix subset are processed, and a CPB band energy feature vector is constructed according to the energy distribution of each CPB band. The element values ​​of the feature vector are energy distribution density values ​​or correlation values ​​with the distribution interval of the main data. The second data acquisition module is used to acquire the second sensor monitoring data of the product under test in the current operating state, perform Fourier transform on the second sensor monitoring data to obtain the CPB spectrum of the second sensor monitoring data, divide the CPB spectrum of the second sensor monitoring data into several CPB frequency bands, and calculate the energy distribution of the second sensor monitoring data in each CPB frequency band to obtain the feature vector corresponding to the second sensor monitoring data. The CPB band feature comparison module is used to compare the feature vector corresponding to the monitoring data of the second sensor with the CPB band energy feature vector constructed in step S3 to determine the current operating status category of the product under test.

Images