Multi-source information fusion equipment fault monitoring device and state evaluation method

Abstract

The invention belongs to the technical field of fault monitoring, and relates to a multi-source information fusion equipment fault monitoring device and a state evaluation method. According to the method, the possible abrasion problem of equipment is deduced from the time-varying characteristics of multiple information sources of equipment vibration signals, abrasive particle characteristics and lubricating oil physical and chemical characteristics, the time-varying rule in an equipment tribology system is researched, the equipment abrasion health state evolution process and the dynamic evolution mechanism are disclosed, and the equipment abrasion health state can be determined under the real working condition. A method of organically combining multi-source monitoring information feature fusion and an equipment friction wear mechanism is adopted, and a mechanical equipment comprehensive health state evaluation platform is developed, so that effective identification of performance degradation behaviors of multiple types of mechanical equipment is realized, and a maintenance decision is reasonably made by deducing the change condition of reliability. Reliability and safety of full-life-cycle service are guaranteed, and evaluation and early warning of equipment abrasion related faults are achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of fault monitoring, and relates to a multi-source information fusion equipment fault monitoring device and a state evaluation method. Background Art

[0002] The working conditions of equipment in actual applications are complex, and the common single signal source collection method is difficult to effectively and timely detect abnormal status of equipment. How to effectively combine feature information of different dimensions is an important difficulty in achieving accurate evaluation of equipment.

[0003] Multi-source information wear state intelligent assessment technology is gradually becoming a trend of popularization and promotion in the industrial manufacturing industry. Multi-source information can obtain a variety of data from multiple sensors and monitoring equipment, including equipment vibration signals, lubricant wear particle images, lubricant physical and chemical characteristics, etc., providing the ability of information collaborative processing, reducing false alarm rates and missed alarm rates, and realizing comprehensive perception and real-time monitoring of equipment wear status.

[0004] However, the research on intelligent assessment of the wear status of multi-source information equipment still faces challenges. The patent document "202210720237.1" discloses "a method and system for assessing the operating health of a fan bearing based on multi-source data", which mainly performs noise reduction and feature extraction on lubricating grease, white corrosion crack risk, preload, embedded temperature and vibration monitoring data sources, and inputs the extracted multi-source data into a health assessment model based on a BP neural network, outputs the comprehensive health index of the fan main bearing and compares it with the calibration threshold interval, and divides it into four intervals of excellent, good, medium and poor for health assessment. The multi-source data on which this method is based have similar levels and frequencies, and only performs noise reduction on the monitoring data source. It is not suitable for heterogeneous and differential data sources, which affects the effective integration and use of data. In addition, the interference of irrelevant feature information is not considered, and the accuracy of the status assessment cannot be guaranteed. Summary of the invention

[0005] In view of this, the present invention provides an equipment fault monitoring device and a state assessment method with multi-source information fusion, which infers possible wear problems of the equipment from the time-varying characteristics of multiple information sources such as equipment vibration signals, lubricating oil wear particle images and lubricating oil physical and chemical characteristics, studies the time-varying laws in the equipment tribology system, and reveals the evolution process and dynamic evolution mechanism of the equipment wear health state.

[0006] To achieve the above object, the present invention adopts the following technical solution: a multi-source information intelligent evaluation method for equipment wear status, comprising the following steps:

[0007] S1: Multi-source signal monitoring;

[0008] Without stopping the machine, a monitoring platform is built to collect equipment vibration signals, lubricant wear particle images, and lubricant physical and chemical characteristic signals;

[0009] S2: Multi-source information feature extraction and processing;

[0010] Feature extraction models are constructed based on the collected vibration signals, lubricant wear particle images, and lubricant physical and chemical characteristics (viscosity, temperature, and moisture).

[0011] S3: Multi-source information fusion;

[0012] The adaptive fusion kernel regression algorithm is used to adaptively weight the features of the three heterogeneous information sources collected, compress the multi-dimensional feature vector at the same time point into a one-dimensional feature vector, extract the health factor, and construct the health space vector;

[0013] S4: Establish status assessment indicators;

[0014] The sliding window method is used to perform local linear fitting of feature data, which can capture the changes in features from any locally divided time series. The sliding window method divides the health factor time series curve after multi-source information fusion into multiple short intervals, which are four stages: healthy state, early wear, severe wear and failure, thereby realizing the description of the equipment wear evolution process and the classification of health status.

[0015] Further, the specific method of step S2 is:

[0016] S21: The original vibration signal is processed using a cyclic sparse decomposition model to obtain a limited number of envelope spectra, the harmonic noise ratio of each envelope spectrum is calculated, and a feature model is established with 14 time domain, frequency domain and 4 information entropy indicators;

[0017] S22: De-noise the lubricant wear particle monitoring images of the equipment, segment the wear particle chain images, extract multi-scale state features such as wear particle concentration, size, contour, and texture change, and establish a lubricant wear particle feature extraction model based on the wear particle image recognition and classification method based on transfer learning according to the time-varying law of wear particle multi-features;

[0018] S23: The multi-factor coupled decay characteristics of lubricating oil viscosity, temperature and moisture are used to correlate with the wear status of equipment and a lubricating oil decay feature extraction model is established.

[0019] Further, the specific method of step S3 is:

[0020] Assume that the optimal feature set of the equipment is D, where the number of features is K, the length of the full life cycle of each feature is N, F is the feature vector, and the multidimensional feature vector D composed of this feature set is N×K :

[0021]

[0022] The extracted multi-source feature set is adaptively weighted to convert the multi-dimensional feature vector D at the same time point into N×K Compressed into a one-dimensional feature vector :

[0023]

[0024] Then the compressed one-dimensional feature vector Construct the health space feature vector, use a set of equipment life cycle features at the beginning of a part, select the first 1 / N feature vectors for combination, and select the first h features to construct the health space feature vector H h :

[0025]

[0026] For the observed feature vector F at the same time t o and the health space eigenvector H h The corresponding Euclidean distance d j (F t o ,H(j)):

[0027] d j (F t o ,H(j))=|F t o -H(j)|(j=1,2,…,h) (20)

[0028] Use Gaussian weight function to estimate the weight ω of the training sample during reconstruction j :

[0029]

[0030] Where q is the bandwidth parameter of the kernel function. After weighting the training data according to the sample distance, we get F t o The weighted average reconstruction value of The health index HI(t) constructed by the Euclidean distance between the two is:

[0031]

[0032] Further, the specific method of step S4 is:

[0033] Let {x l} is a time series signal with a length of l on the time axis, and the time point in space is defined as The intervals separated by the sliding window are:

[0034] X i =(x(t1+s(i-1) ),x(t 2+s(i-1) ),...x(t w+s(i-1) )) (twenty three)

[0035] Where w is the window length and s is the step size.

[0036] A fault monitoring device used in a multi-source information intelligent evaluation method for equipment wear state, comprising a vibration signal monitoring module, a lubricating oil wear particle monitoring module, a lubricating oil physical and chemical characteristics monitoring module, a peristaltic pump, a single chip microcomputer and a host computer;

[0037] The vibration signal monitoring module includes a wired vibration sensor and a wireless vibration sensor; the wired vibration sensor is arranged at the center of the X-axis of the mechanical equipment box, and the wireless vibration sensor is arranged at the center of the Y-axis of the mechanical equipment box, and the vibration signal is synchronously collected by the wired vibration sensor and the wireless vibration sensor;

[0038] The lubricating oil wear particle monitoring module includes an image acquisition sensor, a light, and a coil;

[0039] The lubricating oil physical and chemical characteristics monitoring module includes a viscosity sensor, a temperature sensor and a moisture sensor;

[0040] The peristaltic pump draws lubricating oil through the oil inlet of the mechanical equipment, passes through the flow channel, flows through the lubricating oil wear particle monitoring module and the lubricating oil physical and chemical characteristics monitoring module, and is connected to the oil outlet of the mechanical equipment; in the process of the lubricating oil passing through the flow channel, the wear particles in the lubricating oil are magnetically adsorbed by the coil, and the lubricating oil wear particle image is collected by the image acquisition sensor. The lubricating oil flowing out of the channel enters the lubricating oil physical and chemical characteristics monitoring module, and the lubricating oil flowing out returns to the equipment lubrication system from the upper end of the mechanical equipment. The single-chip microcomputer controls the working parameters of the peristaltic pump to extract lubricating oil, light, and magnetic field coil. The host computer and the single-chip microcomputer are connected by an RS485 serial communication interface. The single-chip microcomputer transmits the collected data to the host computer. During the operation of the equipment, the lubricating oil wear particle image and physical and chemical indicators are collected in real time to realize a series of operations such as data visualization.

[0041] Furthermore, the model of the wired vibration sensor is CT9204; the model of the wireless vibration sensor is CardVibroAir2.

[0042] Compared with the prior art, the advantages of the present invention are:

[0043] 1. The present invention adopts a method of fusing time-varying features of multiple information sources such as equipment vibration signals, wear particle characteristics and physical and chemical characteristics of lubricating oil. Compared with existing evaluation methods, it monitors multi-source information of a wider range of levels and frequencies, more effectively characterizes the working status and health status of the equipment, improves the ability of information collaborative processing, and helps to reduce false alarm rates and missed alarm rates.

[0044] 2. The present invention adopts multi-source information feature extraction and processing, which can well reflect the severity of the fault, has a high degree of surface sensitivity, effectively ensures the robustness of the model, and prevents interference from irrelevant feature information.

[0045] 3. The present invention uses an adaptive multi-source information fusion kernel regression algorithm to adaptively weight the characteristics of three heterogeneous information sources, compress the multi-dimensional feature vector at the same time point into a one-dimensional feature vector, extract the health factor, correct the deviation through the historical observation feature vector of the equipment, and construct a healthy space vector; this method can relatively quickly realize the compression of the feature matrix on the time axis, reduce the complexity of calculation, and only require the feature vector in the healthy state when constructing the healthy space vector, and can realize the correction of abnormal data without the characteristics of the degradation period.

[0046] 4. The method of the present invention can effectively identify the performance degradation behavior of multiple categories of equipment, make reasonable maintenance decisions by inferring the changes in reliability, ensure the reliability and safety of service throughout the life cycle, and realize the evaluation and early warning of equipment wear-related failures.

[0047] 5. The present invention uses multiple signal sources and multiple data mining methods to provide diversified information sources, thereby extracting the original characteristics of the signal more comprehensively and accurately, and improving the overall accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of an intelligent evaluation method for equipment wear status based on multi-source information is invented;

[0049] Figure 2 This is a schematic diagram of the structure of the lubricating oil monitoring system of the present invention;

[0050] Figure 3 It is a time domain waveform diagram and envelope spectrum diagram of the vibration signal of the mechanical equipment of the present invention;

[0051] Figure 4 The image of lubricating oil wear particles of the mechanical equipment of the present invention;

[0052] Figure 5 The physical and chemical linear relationship of the lubricating oil for mechanical equipment of the present invention;

[0053] Figure 6 It is divided into the wear states of the equipment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below in conjunction with the drawings.

[0055] This embodiment provides an intelligent fault monitoring device for equipment wear status with multi-source information, such as Figure 2 As shown, it includes equipment vibration signal monitoring module, lubricating oil wear particle monitoring module, lubricating oil physical and chemical characteristics monitoring module, peristaltic pump, single chip microcomputer and host computer part.

[0056] The vibration signal monitoring module includes a wired vibration sensor CT9204, which is set at the center of the X-axis of the mechanical equipment box; a wireless vibration sensor CardVibro Air2, which is set at the center of the Y-axis of the mechanical equipment box, and the upper computer is equipped with supporting software to synchronously collect vibration signals.

[0057] A threaded hole is set at the bottom of the mechanical equipment lubrication system as an oil outlet, and an external two-way pipe is connected to a peristaltic pump to extract lubricating oil to the lubricating oil monitoring system. The lubricating oil monitoring system includes a lubricating oil wear particle monitoring module and a lubricating oil physical and chemical characteristics monitoring module.

[0058] The lubricating oil wear particle monitoring module includes image acquisition sensors, lights, coils and other parts. When the lubricating oil passes through the flow channel, the wear particles in the lubricating oil are magnetically adsorbed by the coil, and the lubricating oil wear particle image is collected by the image acquisition sensor. The lubricating oil flowing out of the channel enters the lubricating oil physical and chemical characteristics monitoring module, which contains a viscosity sensor, a temperature sensor and a moisture sensor. The outflowing lubricating oil returns to the equipment lubrication system from the upper end of the mechanical equipment. The single-chip microcomputer system controls the working parameters of the peristaltic pump to extract the lubricating oil, the light, and the magnetic field coil. The host computer and the single-chip microcomputer are connected by an RS485 serial communication interface. The single-chip microcomputer transmits the collected data to the host computer. During the operation of the equipment, the lubricating oil wear particle image and physical and chemical indicators are collected in real time to realize a series of operations such as data visualization.

[0059] An intelligent evaluation method for equipment wear status based on multi-source information, such as Figure 1 As shown, the following steps are included:

[0060] S1: Multi-source signal monitoring device;

[0061] Through the above-mentioned intelligent fault monitoring device, the vibration signal, wear particle image and physical and chemical characteristics of the lubricating oil of the equipment can be collected without stopping the machine.

[0062] S2: Multi-source information feature extraction and processing;

[0063] Feature extraction models are constructed based on the vibration signals, lubricant wear particle images, and lubricant physical and chemical characteristics (viscosity, temperature, and moisture) collected during the operation of the equipment.

[0064] S21: Use the cyclic sparse algorithm to process the original signal and obtain a limited number of envelope spectra. Calculate the harmonic noise ratio of each envelope spectrum, and construct a feature vector with 14 time domain and frequency domain (minimum value, frequency root mean square, margin factor, clearance factor, pulse factor, kurtosis, skewness, maximum value, rectified average value, standard deviation, peak factor, peak value, waveform factor, mean) and 4 information entropy (singular value entropy, permutation entropy, approximate entropy, power spectrum entropy) indicators, such as Figure 3 shown.

[0065] First, the sample is enveloped, and then the ratio between the harmonic energy and the noise energy is calculated, which can well reflect the severity of the fault and has high sensitivity and robustness. Without prior knowledge, the envelope harmonic-to-noise ratio after cyclic sparse decomposition is widely used in adaptive extraction of fault features.

[0066] Assume x(t) has n q samples, each sample has m q Sample data, perform Hilbert transform on x(t) to get ,

[0067]

[0068] right Calculate the envelope and remove the DC component to get a new time series

[0069]

[0070] Where e(t) is the time series after envelope processing, is the mean.

[0071] Calculate the autocorrelation function r of the new time series x (τ):

[0072]

[0073] Where ζ is the time lag of the autocorrelation function.

[0074] Calculate the envelope harmonic-to-noise ratio EHNR:

[0075]

[0076] In the formula, ζ max For rx (ζ) maximum position of the autocorrelation function; r x (ζ max ) is the amplitude at the maximum position; r x (0) is the total energy of the envelope spectrum.

[0077] 14 time domain and frequency domain statistical characteristic parameters are calculated, as shown in Table 1 below:

[0078] Table 1 shows the characteristic parameters in the time-frequency domain.

[0079]

[0080]

[0081] (Note: x is the vibration signal point, f k is the frequency amplitude of the signal point, P(k) is the power spectrum value, and N is the data length)

[0082] The calculation of the four information entropy statistical characteristic parameters is shown in Table 2 below:

[0083] Table 2 shows the information entropy characteristic parameters.

[0084]

[0085] S22: The monitoring images of lubricating oil wear particles of the equipment were de-noised, the wear particle chain images were segmented, and the multi-scale state characteristics such as wear particle concentration, size, contour, and texture change were extracted. According to the time-varying law of wear particle multi-features, a wear particle image recognition and classification method based on transfer learning was proposed, and a lubricating oil wear particle feature extraction model was established, such as Figure 4 shown.

[0086] like Figure 4 In Figure 2, the wear information in the wear image has a certain structure and can be expressed by a linear combination of atoms. Therefore, when constructing the sparse components of the wear image, the wear information can be accurately distinguished from the noise, and then the clean image can be reconstructed using the constructed sparse components to obtain the denoised image Z:

[0087]

[0088] In the formula, F is the threshold, I is the unit matrix, R ij is the ijth sub-image block.

[0089] The denoised image is grayed, the grayed image is segmented using the Bayesian threshold algorithm, the segmented image is morphologically processed, the foreground and background of the morphologically processed image are marked, and then a modified gradient image is obtained. The above steps are repeated for the modified gradient image, and finally the result image is segmented by watershed. Figure 4Figure 3, the first improved morphological gradient image g WS (m,n):

[0090] g WS (m,n)=m min[g(m,n),g imark (m,n)|g emark (m,n)] (6)

[0091] Where mmin[] is the minimum imposition operation, g imark (m,n) is the foreground, e emark (m,n) is the background, and g(m,n) is the morphological gradient image.

[0092] The wear mechanisms of the friction pairs of various equipment in contact with each other are different, resulting in different forms and severity of wear, such as Figure 4 Figure 4 shows the different wear types and wear particles, as shown in Table 3 below:

[0093] Table 3 shows 5 common abrasives

[0094]

[0095] like Figure 4 In Figure 5, the IPCA of the wear particle concentration in the wear particle image can directly reflect the change in the wear amount of the equipment:

[0096]

[0097] In the formula, ∑ψ i is the number of pixels of all wear particles in the image, W w is the overall image height, H h is the overall image width.

[0098] S23: Use the multi-factor coupled decay characteristics of lubricating oil viscosity, temperature and moisture to correlate with the wear status of equipment and establish a lubricating oil decay feature extraction model, such as Figure 5 shown.

[0099] Relative viscosity is the ratio of the dynamic viscosity of the polymer solution to the dynamic viscosity of the pure solvent at the same temperature, that is, v re =v ru / v oil . Relative viscosity v re The linear relationship with temperature and moisture content is:

[0100] v re =1+θ(T)φ (8)

[0101] Where θ(T) is a function of temperature T, that is, the slope of the curve of relative viscosity changing with water content, and φ is the water content.

[0102] Since θ(T) and φ are both very small compared to 1, the logarithmic form of both sides is as follows:

[0103]

[0104] Andrade viscosity-temperature equation v oil =A'e B' / T Substituting in:

[0105]

[0106] In the formula, v ru is the kinematic viscosity of the emulsified oil, η0 is the initial dynamic viscosity, and ρ is the density of the lubricating oil.

[0107] A set of viscosity and temperature time series signals {v(n)|n=0,1,...,N-1}, {u(n)|n=0,1,...,N-1} of length N are collected, and polynomial linear fitting is performed to solve the linear equation group calculation value G(n) and residual value L(n), as well as time {t(n)|n=0,1,...,N-1}.

[0108]

[0109] Due to the necessary conditions for the extreme value of a binary function, let Right now:

[0110]

[0111] The viscosity fitting result V = aT + b is obtained, and the temperature fitting result U = aT + b. Due to the influence of temperature on the rate of change of viscosity in time series, the S1 characteristic index is proposed to represent its change process.

[0112] By finding the difference d[V(i)] and d[U(i)] between the corresponding elements of the two vectors V(i) and U(i), the viscosity-temperature characteristic S1 is obtained as follows:

[0113]

[0114] S3: Multi-source information fusion;

[0115] The adaptive fusion kernel regression algorithm is used to adaptively weight the features of three heterogeneous information sources, compress the multi-dimensional feature vector at the same time point into a one-dimensional feature vector, extract the health factor, and construct the health space vector.

[0116] Assume that the optimal feature set of the equipment is D, where the number of features is K, the length of the full life cycle of each feature is N, F is the feature vector, and the multidimensional feature vector D composed of this feature set is N×K :

[0117]

[0118] The extracted multi-source feature set is adaptively weighted to convert the multi-dimensional feature vector D at the same time point into N×K Compressed into a one-dimensional feature vector

[0119]

[0120] Then the compressed one-dimensional feature vector Construct the health space feature vector, use a set of equipment life cycle features at the beginning of a part, select the first 1 / N feature vectors to combine, select the first h features to construct the health space feature vector H h :

[0121] H h =[F1 o ,F2 o ,…,F h o ]=[F1,F2,…,F h ] (19)

[0122] For the observed feature vector F at the same time t t o and the health space eigenvector H h The corresponding Euclidean distance d j (F t o ,H(j)):

[0123] d j (F t o ,H(j))=|F t o -H(j)|(j=1,2,…,h) (20)

[0124] Use Gaussian weight function to estimate the weight ω of the training sample during reconstruction j :

[0125]

[0126] Where q is the bandwidth parameter of the kernel function. After weighting the training data according to the sample distance, we get F t o The weighted average reconstruction value of The health index HI(t) constructed by the Euclidean distance between the two is:

[0127]

[0128] S4: Establish status assessment indicators;

[0129] The sliding window method is used to perform local linear fitting of feature data, which can capture the changes of features from any local divided time series. The sliding window method divides the health factor time series curve after multi-source information fusion into multiple short intervals, which are four stages: healthy state, early wear, severe wear and failure, to achieve the description of equipment wear evolution process and health state classification, such as Figure 6 shown.

[0130] Let {x l} is a time series signal with a length of l on the time axis, and the time point in space is defined as The intervals separated by the sliding window are as follows:

[0131] X i =(x(t 1+s(i-1) ),x(t 2+s(i-1) ),…x(t w+s(i-1) )) (twenty three)

[0132] The above method of the present invention can solve the problem of single-source signal fault warning accuracy of existing equipment. Since the analysis of the physical and chemical characteristic signal of lubricating oil is mainly low-frequency analysis, in contrast, the frequency range of vibration signal is usually wider and can provide more detailed wear information, and the wear particle image can provide wear state classification.

[0133] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent assessment method for equipment wear status based on multi-source information. It is characterized in that The following steps are involved: S1: Multi-source signal monitoring; Without stopping the machine, a monitoring platform is built to collect equipment vibration signals, lubricant wear particle images, and lubricant physical and chemical characteristic signals; S2: Multi-source information feature extraction and processing; Feature extraction models are constructed based on the collected vibration signals, lubricant wear particle images, and lubricant physical and chemical characteristics (viscosity, temperature, and moisture). S3: Multi-source information fusion; The adaptive fusion kernel regression algorithm is used to adaptively weight the features of the three heterogeneous information sources collected, compress the multi-dimensional feature vector at the same time point into a one-dimensional feature vector, extract the health factor, and construct the health space vector; S4: Establish status assessment indicators; The sliding window method is used to perform local linear fitting of feature data, which can capture the changes in features from any locally divided time series. The sliding window method divides the health factor time series curve after multi-source information fusion into multiple short intervals, which are four stages: healthy state, early wear, severe wear and failure, thereby realizing the description of the equipment wear evolution process and the classification of health status.

2. According to the multi-source information intelligent assessment method for equipment wear status according to claim 1, Features: The specific method of step S2 is: S21: The original vibration signal is processed using a cyclic sparse decomposition model to obtain a limited number of envelope spectra, the harmonic noise ratio of each envelope spectrum is calculated, and a feature model is established with 14 time domain, frequency domain and 4 information entropy indicators; S22: De-noise the lubricant wear particle monitoring images of the equipment, segment the wear particle chain images, extract multi-scale state features such as wear particle concentration, size, contour, and texture change, and establish a lubricant wear particle feature extraction model based on the wear particle image recognition and classification method based on transfer learning according to the time-varying law of wear particle multi-features; S23: The multi-factor coupled decay characteristics of lubricating oil viscosity, temperature and moisture are used to correlate with the wear status of equipment and a lubricating oil decay feature extraction model is established.

3. The intelligent evaluation method for equipment wear status based on multi-source information according to claim 1 or 2, Features: The specific method of step S3 is: Assume that the optimal feature set of the equipment is D, where the number of features is K, the length of the full life cycle of each feature is N, F is the feature vector, and the multidimensional feature vector D composed of this feature set is N×K : The extracted multi-source feature set is adaptively weighted to convert the multi-dimensional feature vector D at the same time point into N×K Compressed into a one-dimensional feature vector Then the compressed one-dimensional feature vector Construct the health space feature vector, use a set of equipment life cycle features at the beginning of a part, select the first 1 / N feature vectors for combination, and select the first h features to construct the health space feature vector H h : For the observed feature vector F at the same time t o and the health space eigenvector H h The corresponding Euclidean distance Use Gaussian weight function to estimate the weight ω of the training sample during reconstruction j : Where q is the bandwidth parameter of the kernel function. After weighting the training data according to the sample distance, we get F t o The weighted average reconstruction value of The health index HI(t) constructed by the Euclidean distance between the two is:

4. According to claim 3, a multi-source information intelligent assessment method for equipment wear status, Features: The specific method of step S4 is: Let {x l } is a time series signal with a length of l on the time axis, and the time point in space is defined as The intervals separated by the sliding window are: X i =(x(t 1+s(i-1) ),x(t 2+s(i-1) ),...x(t w+s(i-1) )) (23) Where w is the window length and s is the step size.

5. The fault monitoring device used in the intelligent evaluation method for equipment wear status based on multi-source information according to claim 1, Features: It includes a vibration signal monitoring module, a lubricating oil wear particle monitoring module, a lubricating oil physical and chemical characteristics monitoring module, a peristaltic pump, a single chip microcomputer and a host computer; The vibration signal monitoring module includes a wired vibration sensor and a wireless vibration sensor; the wired vibration sensor is arranged at the center of the X-axis of the mechanical equipment box, and the wireless vibration sensor is arranged at the center of the Y-axis of the mechanical equipment box, and the vibration signal is synchronously collected by the wired vibration sensor and the wireless vibration sensor; The lubricating oil wear particle monitoring module includes an image acquisition sensor, a light, and a coil; The lubricating oil physical and chemical characteristics monitoring module includes a viscosity sensor, a temperature sensor and a moisture sensor; The peristaltic pump draws lubricating oil through the oil inlet of the mechanical equipment, passes through the flow channel, flows through the lubricating oil wear particle monitoring module and the lubricating oil physical and chemical characteristics monitoring module, and is connected to the oil outlet of the mechanical equipment; in the process of the lubricating oil passing through the flow channel, the wear particles in the lubricating oil are magnetically adsorbed by the coil, and the lubricating oil wear particle image is collected by the image acquisition sensor. The lubricating oil flowing out of the channel enters the lubricating oil physical and chemical characteristics monitoring module, and the lubricating oil flowing out returns to the equipment lubrication system from the upper end of the mechanical equipment. The single-chip microcomputer controls the working parameters of the peristaltic pump to extract lubricating oil, light, and magnetic field coil. The host computer and the single-chip microcomputer are connected by an RS485 serial communication interface. The single-chip microcomputer transmits the collected data to the host computer. During the operation of the equipment, the lubricating oil wear particle image and physical and chemical indicators are collected in real time to realize a series of operations such as data visualization.

6. The fault monitoring device used in the intelligent evaluation method for equipment wear status based on multi-source information according to claim 5, Features: The model of the wired vibration sensor is CT9204; the model of the wireless vibration sensor is CardVibro Air2.

Images