A method and system for online detection of a traction wheel

Abstract

The present application belongs to the technical field of automobile parts detection, and particularly relates to a kind of tensioner online detection method and system.The system includes data acquisition subsystem, online analysis and diagnosis subsystem and multi-stage warning subsystem;Data acquisition subsystem synchronously acquires tensioner working condition data through multi-source sensor array, online analysis and diagnosis subsystem fuses latent variable feature extraction, blind source separation and independent component analysis functions, realizes tensioner health state detection, fault accurate identification and positioning by constructing frequency constraint, envelope constraint and similarity constraint, and can also simultaneously predict residual life.The present application solves the defects of low efficiency of traditional offline detection, inability to reflect real working state and insufficient fault positioning accuracy, realizes real-time detection of tensioner under multiple working conditions, has high detection accuracy and strong robustness, and is suitable for full life cycle monitoring of automobile engine tensioner.

Description

Technical Field

[0001] This invention belongs to the field of automotive parts testing technology, specifically relating to an online condition detection method and system for tensioners. Background Technology

[0002] The tensioner pulley is one of the core components of the automotive engine transmission system and a key component to ensure the normal operation of the engine. The tensioner pulley provides stable and adjustable tension to the belt and chain, which can compensate for the looseness caused by thermal expansion and contraction, elastic deformation and long-term wear, and is used to eliminate belt slippage, jumping and abnormal noise.

[0003] Currently, the inspection of traditional tensioners mainly relies on offline, single-parameter testing methods. This involves manually inspecting the tensioner's exterior for cracks, bearing oil leaks, tension force, and loose bolts to make a preliminary assessment of its health. Traditional methods suffer from low efficiency, inability to reflect the tensioner's true working condition in real time, and poor predictability of tensioner failures.

[0004] Therefore, there is an urgent need for a method to overcome the shortcomings of existing detection technologies in the detection and prediction of tensioner failures. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an online detection method and system for tensioners. This invention integrates latent variable feature extraction and independent component analysis techniques to achieve real-time monitoring, accurate fault identification and location, and remaining life prediction of the tensioner under operating conditions, forming a closed-loop detection system.

[0006] This invention proposes an online tensioner detection system, comprising a data acquisition subsystem, an online analysis and diagnosis subsystem, and a multi-level alarm subsystem. The data acquisition subsystem is communicatively connected to the online analysis and diagnosis subsystem, and the online analysis and diagnosis subsystem is communicatively connected to the multi-level alarm subsystem. This system is adapted to the online detection requirements of automotive engine tensioners.

[0007] The data acquisition subsystem includes a multi-source sensor array and a data transmission unit. The multi-source sensor array consists of vibration sensors, an infrared thermal imager, and noise sensors. The vibration sensors are mounted on the tensioner bearing housing, prioritizing axial and radial measurements to differentiate between different types of faults. The infrared thermal imager is positioned directly opposite the outer ring of the bearing, periodically scanning to obtain the temperature field distribution of the tensioner. The noise sensors are pointed towards the tensioner area and equipped with dustproof devices to collect high-frequency abnormal noise signals. The data transmission unit transmits the data acquired by the sensors to the online analysis and diagnostic subsystem.

[0008] The data acquisition subsystem uses a multi-source sensor array to synchronously identify the working conditions of the tensioner. Each sensor synchronously locks onto corresponding data information (vibration, temperature, noise) through mutual sensing. The processor determines whether preset working condition information has been identified. If preset working condition information is identified, the data acquisition window is triggered to collect and record the corresponding multi-source key data. If preset working condition information is not identified, the synchronous identification of working conditions continues. The processor outputs the synchronous multi-source dataset to the online analysis and diagnosis subsystem.

[0009] The online analysis and diagnosis subsystem includes a data preprocessing module, a feature extraction module, a dimensionality reduction module, a blind source separation module, an independent component analysis module, and a hierarchical diagnosis module.

[0010] The online analysis and diagnosis subsystem receives multi-source datasets collected by the data acquisition subsystem. The data preprocessing module processes the multi-source datasets, including anomaly labeling, noise filtering, operating condition alignment, and normalization. The feature extraction module extracts features and integrates them into a high-dimensional feature vector. The dimensionality reduction module reduces the dimensionality of the high-dimensional feature vector to obtain a core feature set. The blind source separation module uses FastICA, JADE, and SOBI algorithms to extract independent components, and after preliminary screening, obtains fault-related components. The independent component analysis module constructs triple constraints, optimizes the unmixing matrix, and reconstructs the signal to obtain the fault-dominant component. The hierarchical diagnosis module performs a first-level health diagnosis of the tensioner pulley using the core feature set and the fault-dominant component, determining whether the first-level diagnosis result is abnormal. If the health status is normal, continuous data monitoring continues. If the health status is abnormal, second-level fault diagnosis, third-level fault location, and fourth-level remaining life prediction diagnosis are performed, outputting the tensioner pulley detection, location, and prediction results to the multi-level alarm subsystem.

[0011] This invention also proposes an online detection method for tensioners, the method comprising the following steps: Step S1: Synchronously identify the working condition of the target tensioner wheel through a multi-source sensor array, lock the data information of each sensor, trigger the data acquisition window to collect data, and output a synchronous multi-source dataset. Step S2 involves performing outlier labeling, noise filtering, operating condition alignment, and normalization on the synchronized multi-source dataset. Step S3: Extract multiple features and combine them into a high-dimensional feature vector. Reduce the dimensionality of the high-dimensional feature vector and retain the core feature set. Step S4: The FastICA, JADE and SOBI algorithms are used to perform blind source separation on the multi-channel vibration signal, extract independent components, and obtain fault-related components after preliminary screening. Step S5: Construct frequency constraints, envelope constraints, and similarity constraints; optimize the unmixing matrix using independent component analysis algorithm and reconstruct the signal to obtain the fault-dominant component. Step S6: Perform a first-level health diagnosis of the tensioner using the core feature set and the fault-dominant component to determine if the first-level diagnosis result is abnormal. If the health status is normal, continue with the continuous data monitoring operation. If the health status is abnormal, perform a second-level fault diagnosis, a third-level fault location, and a fourth-level remaining life prediction diagnosis.

[0012] Compared with the prior art, the present invention has the following advantages: This invention enables real-time monitoring of the tensioner's working state. By synchronously collecting tensioner condition data through a multi-source sensor array and combining it with online analysis and diagnosis, the invention reflects the true working state of the tensioner in real time, overcoming the shortcomings of traditional offline detection methods that cannot monitor in real time.

[0013] High diagnostic accuracy and precise location; it integrates latent variable feature extraction, blind source separation and multi-constraint independent component analysis techniques, and reconstructs the dominant fault component through triple constraints to accurately identify the fault type and locate the fault location with a location accuracy of ≥90%, solving the problem of fuzzy location in traditional methods. It has strong adaptability to various working conditions; through optimization of time-frequency domain feature extraction and independent component analysis algorithms, it can adapt to complex industrial working conditions such as constant speed, variable speed, and multiple loads, and has strong anti-interference ability. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a block diagram of an online detection system based on a tensioner pulley proposed in this invention; Figure 2 This is a schematic diagram of the data acquisition subsystem of the present invention; Figure 3 This is a schematic diagram of the data acquisition subsystem of the present invention. Figure 4 This is a flowchart of an online detection method for tensioners proposed in this invention. Detailed Implementation

[0016] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0017] like Figure 1 As shown, this embodiment of the invention provides an online detection system based on a tensioner, including a data acquisition subsystem 1, an online analysis and diagnosis subsystem 2, and a multi-level alarm subsystem 3. The data acquisition subsystem 1 is communicatively connected to the online analysis and diagnosis subsystem 2, and the online analysis and diagnosis subsystem 2 is communicatively connected to the multi-level alarm subsystem 3. This system is adapted to the online detection requirements of the tensioner 4 of an automotive engine.

[0018] like Figure 2 As shown, the data acquisition subsystem 1 includes a multi-source sensor array and a data transmission unit. The multi-source sensor array consists of a vibration sensor 11, an infrared thermal imager 12, and a noise sensor 13. The vibration sensor 11 uses a PCBICP353C03 accelerometer, mounted on the tensioner bearing housing, and preferentially measures axial and radial directions to distinguish different types of faults. The infrared thermal imager 12 faces the outer ring of the bearing and obtains the temperature field distribution of the tensioner through periodic scanning. The noise sensor 13 is pointed towards the tensioner area and equipped with a dustproof device to collect high-frequency abnormal noise signals. The data transmission unit transmits the data collected by the sensors to the online analysis and diagnostic subsystem 2.

[0019] like Figure 3 As shown, the data acquisition subsystem 1 uses a multi-source sensor array to synchronously identify the working condition of the tensioner wheel; each sensor synchronously locks the corresponding data information (vibration, temperature, noise) through mutual sensing; the processor determines whether the preset working condition information is identified; if the preset working condition information is identified, the data acquisition window is triggered to collect and record the corresponding multi-source key data; if the preset working condition information is not identified, the working condition synchronous identification continues; the processor outputs the synchronous multi-source dataset to the online analysis and diagnosis subsystem 2.

[0020] The online analysis and diagnosis subsystem 2 includes a data preprocessing module 21, a feature extraction module 22, a dimensionality reduction module 23, a blind source separation module 24, an independent component analysis module 25, and a hierarchical diagnosis module 26.

[0021] The online analysis and diagnosis subsystem 2 is mainly used to receive the synchronous multi-source dataset transmitted by the data acquisition subsystem 1; the data preprocessing module 21 processes the synchronous multi-source dataset, including abnormal data marking, noise filtering, working condition alignment and normalization; the feature extraction module 22 extracts features and integrates them to form a high-dimensional feature vector; the dimensionality reduction module 23 reduces the dimensionality of the high-dimensional feature vector to obtain the core feature set; the blind source separation module 24 uses three algorithms, FastICA, JADE and SOBI, to extract independent components, and obtains fault-related components after preliminary screening; the independent component analysis module 25 constructs triple constraint conditions, optimizes the unmixing matrix and reconstructs the signal to obtain the fault-dominant component; the hierarchical diagnosis module 26 performs a first-level health diagnosis of the tensioner wheel through the core feature set and the fault-dominant component, and judges whether the first-level diagnosis result is abnormal; if the health status is normal, the continuous data monitoring operation continues; if the health status is abnormal, the second-level fault diagnosis, the third-level fault location and the fourth-level remaining life prediction diagnosis are performed; the tensioner wheel detection, location and prediction results are output to the multi-level alarm subsystem 3.

[0022] The specific processing procedure of the data preprocessing module 21 is as follows: The input synchronous multi-source dataset is parsed into time-series data groups X=[x1,x2,x3,...,xn] for each sensor channel; Calculate the mean μ and standard deviation of the data for each channel. And set a dynamic threshold θ=μ±5 Upper limit of absolute value If |xi|>min( If |θ| is an outlier data point, then it is marked as an outlier data point.

[0023] After filtering the dataset for noise, the target vibration signal e[n] is obtained:

[0024]

[0025] in, Main signal, For reference signal, For noise signals, This is the target signal after noise filtering.

[0026] Select the input operating condition parameter T that satisfies The continuous time interval [t1, t2] is defined, and the synchronization data of all sensor channels is extracted, where The rise rate threshold is used to normalize the data x in a single channel.

[0027] The feature extraction module includes the following steps: Calculate the mean μ, root mean square value Xrms, peak value Xm, variance, , standard deviation , kurtosis K, skewness S, waveform factor SF, peak factor CF, impulse factor IF, and margin factor CLF for the preprocessed discrete signal sequence x[n].

[0028] The blind source separation module 24 processes the multi-channel vibration signals using three algorithms: FastICA, JADE, and SOBI. By comparing the spectra of the independent components extracted by the three algorithms, the components related to the fault characteristic frequencies are preliminarily screened out.

[0029] The independent component analysis module 25 first constructs constraint conditions, including: ① Construct a frequency constraint vector rf with the fault characteristic frequencies (inner race fault frequency BPFI / outer race fault frequency BPFO / roller fault frequency BSF / cage fault frequency FTF) as a reference; ② Conduct envelope analysis on the preliminarily screened fault-related components and construct an envelope constraint vector re with the periodic impact envelope as a reference; ③ Set the constraint thresholds of Jaccard index JI≥0.95 and correlation coefficient CC≥0.99; Use the above triple constraint conditions to solve the optimized demixing matrix through the Newton-like learning algorithm , and reconstruct the fault-dominated independent components through the following formula to eliminate the interference of vibrations from other components and environmental noise.

[0030]

[0031] In the formula, is the multi-channel vibration time series signal after outlier marking, noise filtering, working condition alignment, and normalization. Based on the traditional ICA demixing matrix, frequency constraints, envelope constraints, and similarity constraints are introduced, and the constraint conditions are more in line with the fault characteristics of the tensioner bearing.

[0032] The hierarchical diagnosis module 26 includes first-level health diagnosis, second-level fault diagnosis, third-level fault location, and fourth-level remaining life prediction diagnosis.

[0033] Among them, the first-level health diagnosis is used to quickly determine whether the tensioner is in a healthy working state, distinguish between healthy / abnormal states, and provide a trigger basis for subsequent in-depth diagnosis. It does not distinguish specific fault types and only makes an overall state judgment; First, combine the key feature vector F = [Xrms, Xm, μ, , K, S, CF, CLF, IF, SF]; For each feature in the key feature vector extracted in real time, calculate the deviation degree Zi = |Fi - Bi| / Si (Bi is the health baseline of the corresponding feature, and Si is the standard deviation of the corresponding feature) according to the formula. If all Zi < T (T = 0.8), it is determined to be healthy. When any Zi ≥ T, it is determined to be abnormal and triggers the second-level fault diagnosis.

[0034] Level 2 fault diagnosis is used to accurately identify the specific fault type (bearing inner ring / outer ring / roller / cage fault) based on the abnormal conditions identified in Level 1 diagnosis, and to assess the severity of the fault (mild / moderate / severe), providing a basis for alarm and maintenance. First, the fault characteristic frequencies are calculated, including the inner ring fault frequency (BPFI), outer ring fault frequency (BPFO), roller fault frequency (BSF), and cage fault frequency (FTF). Then, a fast Fourier transform is performed on the preprocessed vibration signal to obtain the frequency domain power spectrum. Extract the spectral centroid FC and spectral variance FV; find the frequency point corresponding to the peak amplitude in the power spectrum. If the deviation of the peak frequency from a certain frequency in BPFI / BPFO / BSF / FTF is ≤±2%, and is verified by FC (the spectral centroid shifts towards the fault characteristic frequency when there is a fault) and FV (the spectral variance increases when there is a fault), it is determined to be a fault of the corresponding component. If multiple peak frequencies match different fault characteristic frequencies, it is determined to be a composite fault, such as a dual fault of inner ring + roller.

[0035] Level 3 fault location, targeting specific faults identified in Level 2 diagnosis, achieves precise positioning from the component level (inner / outer ring) to the point level (e.g., upper side of the inner ring, lower side of the outer ring), providing accurate coordinates for targeted repairs, avoiding full component disassembly, and improving repair efficiency. First, frequency constraints (based on the fault characteristic frequencies of Level 2 diagnosis), envelope constraints (based on the periodic impact envelope of the fault), and similarity constraints are constructed respectively. The dominant fault components are then reconstructed through the independent component analysis module 25. Analyze the energy distribution of fault components in multi-channel sensor signals to locate the fault location.

[0036] Specifically, regarding the dominant components of the fault Perform a Fast Fourier Transform to obtain its spectrum. ,extract main frequency ,like If the deviation from a certain frequency in BPFI / BPFO / BSF / FTF is less than ±2%, it is directly determined that the corresponding component is faulty (inner ring / outer ring / rolling element / cage).

[0037] The Level 4 Remaining Life Predictive Diagnosis, based on the fault type, severity, and precise location results of the first three levels of diagnosis, quantitatively assesses the remaining service life of the tensioner, realizing a shift from "post-failure maintenance" to "predictive maintenance" and rationally planning spare parts and downtime. The core fault-sensitive features extracted from the Level 3 diagnosis (such as kurtosis and fault frequency amplitude) are integrated into a fault severity feature value St, which is normalized to a health index It = exp(- ), It∈[0,1], where 1 is brand new, 0 is completely failed, and S is the characteristic value in the healthy state; calculate the time-weighted average value RUL to obtain the estimated remaining lifetime, and output the judgment result.

[0038] like Figure 4 As shown, the present invention also proposes an online detection method for tensioners, the method comprising the following steps: Step S1: Synchronously identify the working condition of the target tensioner wheel through a multi-source sensor array, lock the data information of each sensor, trigger the data acquisition window to collect data, and output a synchronous multi-source dataset. Step S2 involves performing outlier labeling, noise filtering, operating condition alignment, and normalization on the synchronized multi-source dataset. Step S3: Extract multiple features and combine them into a high-dimensional feature vector. Reduce the dimensionality of the high-dimensional feature vector and retain the core feature set. Step S4: The FastICA, JADE and SOBI algorithms are used to perform blind source separation on the multi-channel vibration signal, extract independent components, and obtain fault-related components after preliminary screening. Step S5: Construct frequency constraints, envelope constraints, and similarity constraints; optimize the unmixing matrix using independent component analysis algorithm and reconstruct the signal to obtain the fault-dominant component. Step S6: Perform a first-level health diagnosis of the tensioner using the core feature set and the fault-dominant component to determine if the first-level diagnosis result is abnormal. If the health status is normal, continue with the continuous data monitoring operation. If the health status is abnormal, perform a second-level fault diagnosis, a third-level fault location, and a fourth-level remaining life prediction diagnosis.

[0039] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0040] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

Claims

1. An online detection system for tensioner pulleys, characterized in that, The detection system includes: a data acquisition subsystem and an online analysis and diagnosis subsystem; The data acquisition subsystem includes a multi-source sensor array and a data transmission unit. The multi-source sensor array consists of a vibration sensor, an infrared thermal imager, and a noise sensor. The vibration sensor is installed on the tensioner bearing housing and measures axially and radially to distinguish different types of faults. The infrared thermal imager is positioned facing the outer ring of the bearing and obtains the temperature field distribution of the tensioner through periodic scanning. The noise sensor is pointed towards the tensioner area and equipped with a dustproof device to collect high-frequency abnormal noise signals. The multi-source sensor array performs synchronous identification of the tensioner's operating conditions, and each sensor synchronously locks onto the corresponding data information through mutual sensing. The processor determines whether preset operating condition information has been identified. If so, the data acquisition window is triggered to collect and record the corresponding multi-source key data; otherwise, synchronous identification of operating conditions continues. The online analysis and diagnosis subsystem receives multi-source datasets collected by the data acquisition subsystem; the data preprocessing module processes the multi-source datasets; the feature extraction module extracts features and integrates them into a high-dimensional feature vector; the dimensionality reduction module reduces the dimensionality of the high-dimensional feature vector to obtain a core feature set; the blind source separation module extracts independent components and obtains fault-related components after preliminary screening; the independent component analysis module constructs triple constraint conditions, optimizes the unmixing matrix, and reconstructs the signal to obtain the fault-dominant component; the hierarchical diagnosis module performs a first-level health diagnosis of the tensioner wheel using the core feature set and the fault-dominant component, and determines whether the first-level diagnosis result is abnormal; if the health status is normal, the continuous data monitoring operation continues; if the health status is abnormal, a second-level fault diagnosis, a third-level fault location, and a fourth-level remaining life prediction diagnosis are performed, and the tensioner wheel detection, location, and prediction results are output to the multi-level alarm subsystem.

2. The online detection system according to claim 1, wherein the independent component analysis module is used to construct triple constraint conditions, specifically: ① constructing a frequency constraint vector rf with fault characteristic frequency as reference; ② performing envelope analysis on the initially screened fault-related components, constructing an envelope constraint vector re with periodic impact envelope as reference; ③ setting constraint thresholds of Jaccard index JI≥0.95 and correlation coefficient CC≥0.

99.

3. The online detection system according to claim 1, wherein the feature extraction specifically comprises calculating the mean μ, root mean square value Xrms, peak value Xm, and variance of the preprocessed discrete signal sequence. Standard deviation The parameters K, S, SF, CF, IF, and CLF are used as eigenvalues.

4. The online detection system according to claim 3, wherein the primary health diagnosis is used to quickly determine whether the tensioner is in a healthy working state and distinguish between two states: healthy / abnormal; first, combine the key feature vector F = [Xrms, Xm, μ, , K, S, CF, CLF, IF, SF]; for each feature in the key feature vector extracted in real time, calculate the deviation degree Zi = |Fi - Bi| / Si, where Bi is the health baseline of the corresponding feature and Si is the standard deviation of the corresponding feature. If all Zi < T and T = 0.8, it is determined to be healthy; when any Zi ≥ T, it is determined to be abnormal and the secondary fault diagnosis is triggered.

5. The online detection system according to claim 1, wherein the secondary fault diagnosis is used to accurately identify the specific fault type based on the abnormal state determined by the primary diagnosis; firstly, the fault characteristic frequencies are calculated, including the inner race fault frequency BPFI, the outer race fault frequency BPFO, the roller fault frequency BSF, and the cage fault frequency FTF; then, a fast Fourier transform is performed on the preprocessed vibration signal to obtain the frequency domain power spectrum. Extract the spectral centroid FC and spectral variance FV; find the frequency point corresponding to the amplitude peak in the power spectrum. If the deviation of the peak frequency from a certain frequency in BPFI / BPFO / BSF / FTF is ≤±2%, and is verified by FC and FV, it is determined to be a fault of the corresponding component; if multiple peak frequencies appear and match different fault characteristic frequencies, it is determined to be a composite fault.

6. The online detection system according to claim 1, wherein the three-level fault localization specifically comprises: constructing frequency constraints, envelope constraints, and similarity constraints respectively, and reconstructing the dominant fault components through an independent component analysis module. Analyze the energy distribution of fault components in multi-channel sensor signals to locate the fault location.

7. The online detection system according to claim 1, wherein the fourth-level remaining service life prediction diagnosis is based on the fault type, severity, and precise location results of the first three levels of diagnosis, and quantitatively assesses the remaining service life of the tensioner; the core sensitive fault features extracted from the third-level diagnosis are fused into a fault severity feature value St, and normalized to a health index It=exp(- ), It∈[0,1], where 1 is brand new, 0 is completely failed, and S is the characteristic value in the healthy state; calculate the time-weighted average value RUL to obtain the estimated remaining lifetime, and output the judgment result.

8. A detection method using the online detection system according to any one of claims 1-7, characterized in that, The method includes the following steps: Step S1: Synchronously identify the working condition of the target tensioner wheel through a multi-source sensor array, lock the data information of each sensor, trigger the data acquisition window to collect data, and output a synchronous multi-source dataset. Step S2 involves performing outlier labeling, noise filtering, operating condition alignment, and normalization on the synchronized multi-source dataset. Step S3: Extract multiple features and combine them into a high-dimensional feature vector. Reduce the dimensionality of the high-dimensional feature vector and retain the core feature set. Step S4: The FastICA, JADE and SOBI algorithms are used to perform blind source separation on the multi-channel vibration signal, extract independent components, and obtain fault-related components after preliminary screening. Step S5: Construct frequency constraints, envelope constraints, and similarity constraints; optimize the unmixing matrix using independent component analysis algorithm and reconstruct the signal to obtain the fault-dominant component. Step S6: Perform a first-level health diagnosis of the tensioner using the core feature set and the fault-dominant component to determine if the first-level diagnosis result is abnormal. If the health status is normal, continue with the continuous data monitoring operation. If the health status is abnormal, perform a second-level fault diagnosis, a third-level fault location, and a fourth-level remaining life prediction diagnosis.

Images