A multidimensional data fault decision method, device and storage medium

Abstract

The present application discloses a multi-dimensional data fault decision method, device and storage medium, which includes: obtaining different monitoring data of equipment, extracting characteristic values; fuzzifying the extracted characteristic values; performing fuzzy reasoning on the processed characteristic values ​​according to a preset fault rule library to generate a decision result; defuzzifying the generated decision result to obtain the final fault decision result and maintenance suggestion. In this way, the objectivity and correctness of the decision result can be guaranteed, the fault information can be fully utilized, the fault location and fault cause can be accurately identified, and maintenance suggestions can be provided in a timely manner.

Description

Technical Field

[0001] The present invention relates to the field of mechanical equipment status monitoring, and in particular to a multi-dimensional data fault decision method, device and storage medium. Background Art

[0002] In the field of mechanical equipment condition monitoring, fault diagnosis of different types of mechanical equipment can be performed based on monitoring data, among which bearing monitoring is the most common. Bearings are key components in all mechanical transmission systems, and their fault status directly affects the working status of the equipment.

[0003] With the rapid development of engines, the working conditions of engine bearings are becoming increasingly harsh and complex. Aircraft engine bearings not only work under high speed, high temperature and high load conditions, but also experience dramatic changes in working conditions. They are prone to failure during operation. Once an aircraft engine bearing fails, it will directly affect the safety of aircraft engines. Some failures occur instantly, while others are caused by slow and long-term degradation. The resulting bearing failures are diverse in form, the severity of the failures caused is also different, and the causes of the failures are not unique.

[0004] The information sources collected by bearing diagnosis are classified, and the diagnosis direction is mainly vibration monitoring and acoustic emission monitoring. Vibration monitoring has many practical applications in rolling bearing fault diagnosis, but the vibration sensor is generally installed far away from the bearing installation position, which has the disadvantages of large signal transmission attenuation, weak signal, and difficulty in characteristic signal extraction. Acoustic emission monitoring has a certain sensitivity to early fault diagnosis, but it is difficult to process and interpret signals, which limits the application of acoustic emission monitoring in mechanical fault diagnosis.

[0005] The currently used engine condition monitoring and fault diagnosis methods not only have single extracted features, but also single reasoning modes, and are unable to fully utilize fault information. There are problems such as difficulty in acquiring fault knowledge, low systematization and intelligence levels, low diagnostic accuracy, and inability to accurately locate the fault location and discover the cause of the fault. Summary of the invention

[0006] In view of this, the purpose of the present invention is to provide a multi-dimensional data fault decision method, device and storage medium, which can ensure the objectivity and correctness of the decision result and accurately identify the fault location and fault cause. The specific scheme is as follows:

[0007] A multidimensional data fault decision method, comprising:

[0008] Obtain different monitoring data of the equipment and extract feature values;

[0009] Performing fuzzy processing on the extracted feature values;

[0010] Perform fuzzy reasoning on the processed characteristic values ​​according to a preset fault rule library to generate a decision result;

[0011] The generated decision result is defuzzified to obtain the final fault decision result and maintenance suggestion.

[0012] Preferably, in the above-mentioned multidimensional data fault decision method provided by an embodiment of the present invention, the characteristic values ​​include any one of frequency domain characteristic values, abnormal sound characteristic values, time domain characteristic values, impact signal characteristic values, abrasive characteristic values, physical and chemical characteristic values, temperature characteristic values, pressure characteristic values, viscosity characteristic values ​​or other characteristic values ​​obtained by mechanism analysis.

[0013] Preferably, in the multidimensional data fault decision method provided by an embodiment of the present invention, when the characteristic value is a time domain characteristic value, extracting the time domain characteristic value includes:

[0014] Extract any of the peak-to-peak value, mean value, RMS value, standard deviation, skewness coefficient, kurtosis, amplitude of the rotor vibration fundamental frequency, and amplitude of twice the rotor vibration fundamental frequency.

[0015] Preferably, in the multidimensional data fault decision method provided by an embodiment of the present invention, when the eigenvalue is a frequency domain eigenvalue, extracting the frequency domain eigenvalue includes:

[0016] The original signal is decomposed using the FIMD algorithm to obtain the intrinsic mode FIMD components;

[0017] Reconstructing the intrinsic mode FIMD component to obtain a reconstructed signal;

[0018] The frequency domain features of the time domain information of the reconstructed signal are extracted by Fourier transform to obtain the signal change trend in the frequency domain, and to extract the information of the frequency band energy, frequency band width and whether there is abnormal frequency.

[0019] Preferably, in the multidimensional data fault decision method provided in the embodiment of the present invention, the original signal is decomposed by using the FIMD algorithm to obtain the intrinsic mode FIMD component, including:

[0020] Calculate the extreme points of the original signal;

[0021] Using a linear transformation method to transform the extreme points of the original signal and generate corresponding residuals;

[0022] The residual amount is subtracted from the original signal to obtain the intrinsic modal FIMD component corresponding to the original signal.

[0023] Preferably, in the above-mentioned multi-dimensional data fault decision method provided by the embodiment of the present invention, the following formula is used to generate the corresponding residual amount:

[0024]

[0025] Where r(t) represents the residual amount, t0, t j and t m-1 Respectively represent the sampling time of different extreme points of the original signal, t j+1 represents the sampling time corresponding to the j+1 extreme value point, R(t j ) represents the sampling time t j The extreme point of R(t j ) represents the sampling time t j+1 The extreme point of .

[0026] Preferably, in the above multi-dimensional data fault decision method provided by the embodiment of the present invention, before subtracting the residual amount from the original signal, it also includes:

[0027] A certain extreme point of the intrinsic mode function is used as a control point for correction;

[0028] The median of the control point is calculated according to the product of two adjacent vectors, and the median of the control point is connected by using a multiple spline connection method to correct the residual amount.

[0029] Preferably, in the multidimensional data fault decision method provided in the embodiment of the present invention, reconstructing the intrinsic mode FIMD component to obtain a reconstructed signal includes:

[0030] Calculate the correlation degree of each order FIMD component;

[0031] The FIMD component corresponding to the correlation degree greater than the set value is selected for reconstruction to obtain a reconstructed signal.

[0032] Preferably, in the multidimensional data fault decision method provided in the embodiment of the present invention, the following formula is used to calculate the correlation degree of each order FIMD component:

[0033]

[0034] in, It indicates the correlation degree of each order FIMD component; represents the cross-correlation function between the FIMD component and the original signal, R S (t) represents the autocorrelation function of the original signal.

[0035] Preferably, in the multidimensional data fault decision method provided by an embodiment of the present invention, when the characteristic value is an impulse signal characteristic value, extracting the impulse signal characteristic value includes:

[0036] Use the resonator to absorb low-frequency fault shocks, generate high-frequency resonance to obtain a free-decay oscillation waveform, and filter out low-frequency vibration signals;

[0037] Performing envelope demodulation on the resonance waveform and performing low-pass filtering on the obtained envelope signal;

[0038] Performing Fourier transform on the envelope signal after low-pass filtering to obtain an envelope spectrum;

[0039] According to the envelope spectrum, any characteristic values ​​of the retainer to the outer ring, the retainer to the inner ring, the outer ring, the inner ring, the roller end face, the roller circumferential surface, the current gear, and the adjacent gear are extracted, and the characteristic value is the corresponding amplitude or dB value on the envelope spectrum.

[0040] Preferably, in the multi-dimensional data fault decision method provided in the embodiment of the present invention, envelope demodulation is performed on the resonance waveform, including:

[0041] The envelope signal is obtained by envelope demodulating the resonance waveform using Hilbert transform or absolute value detection.

[0042] Preferably, in the above-mentioned multidimensional data fault decision method provided by the embodiment of the present invention, while extracting the corresponding amplitude or dB value on the envelope spectrum, it also includes:

[0043] The proportion of the energy of the corresponding characteristic frequency and its multiple frequency in the total energy of the entire envelope spectrum is selected as the impulse signal characteristic value.

[0044] Preferably, in the multidimensional data fault decision method provided in the embodiment of the present invention, the following formula is used to obtain the proportion of the energy of the corresponding characteristic frequency and its multiple frequency in the total energy of the entire envelope spectrum:

[0045]

[0046] in, Indicates the obtained proportion value, Represents the energy sum of the characteristic frequency and its multiples; Represents the energy and frequency of the entire envelope spectrum; y i and j Represents the amplitude at each frequency.

[0047] Preferably, in the multidimensional data fault decision method provided by the embodiment of the present invention, when the characteristic value is a wear particle characteristic value, extracting the wear particle characteristic value includes:

[0048] Identify ferromagnetic and non-ferromagnetic particles through online monitoring of lubricating oil wear particles;

[0049] At the same time, the number and mass of particles within each size range are counted as the abrasive particle characteristic values.

[0050] Preferably, in the multidimensional data fault decision method provided in the embodiment of the present invention, fuzzy processing is performed on the extracted characteristic values, including:

[0051] The eigenvalues ​​are fuzzified by using a membership function.

[0052] Preferably, in the above-mentioned multidimensional data fault decision method provided by the embodiment of the present invention, before fuzzy reasoning is performed on the processed characteristic value according to the preset fault rule library, it also includes:

[0053] Establish relevant rules in the preset fault rule base and perform rule verification;

[0054] If no rule conflict occurs, the rule is stored in the preset fault rule library;

[0055] If there is a rule conflict, the rule or the original rule is modified according to the corresponding knowledge of the fault mechanism, and when there is no rule conflict, the rule is stored in the preset fault rule library.

[0056] Preferably, in the above multi-dimensional data fault decision method provided by the embodiment of the present invention, while establishing relevant rules in the preset fault rule base, it also includes:

[0057] The preset fault rule base is provided with a weight coefficient of a premise condition, a rule activation threshold and a rule credibility to make the rules fuzzy.

[0058] Preferably, in the above-mentioned multi-dimensional data fault decision method provided by the embodiment of the present invention, the weight coefficients of the relevant rules or premise conditions in the preset fault rule base are established based on mechanism analysis.

[0059] Preferably, in the above-mentioned multidimensional data fault decision method provided by the embodiment of the present invention, the following formula is used to set the weight coefficient of the premise condition, the rule activation threshold and the rule credibility of the preset fault rule base:

[0060]

[0061] The above formula is a fuzzy relation expression, where R i represents the i-th inference rule, A i1 , A i2 , …, A in represents the premise of the rule, ω i1 ,ω i2 ,…,ω in is the weight coefficient of the corresponding prerequisite, B i represents the conclusion of the rule, CF represents the credibility of the rule, and τ represents the activation threshold of the rule.

[0062] Preferably, in the above-mentioned multidimensional data fault decision method provided by the embodiment of the present invention, fuzzy reasoning is performed on the processed characteristic values ​​according to a preset fault rule library to generate a decision result, including:

[0063] Performing forward fuzzy reasoning on the processed characteristic values ​​according to a preset fault rule library to generate a decision result;

[0064] If the forward fuzzy reasoning fails, the reverse branch reasoning method is selected to generate the decision result.

[0065] Preferably, in the multidimensional data fault decision method provided in the embodiment of the present invention, forward fuzzy reasoning is performed on the processed feature values ​​according to a preset fault rule base, including:

[0066] storing the processed characteristic value as a fault characteristic fact in a fact database;

[0067] Using a preset matching rule in the preset fault rule base to match the fault feature fact with the rules in the preset fault rule base;

[0068] If the match is successful, the confidence is calculated. If the calculated confidence is greater than or equal to the set threshold, a decision result is generated; if the calculated confidence is less than the set threshold, the forward fuzzy reasoning is judged to have failed.

[0069] Preferably, in the above multidimensional data fault decision method provided by the embodiment of the present invention, after the matching is successful and before the confidence is calculated, it also includes:

[0070] When multiple rules are matched at the same time, the conflicts are resolved according to the preset conflict resolution method.

[0071] Preferably, in the multidimensional data fault decision method provided in the embodiment of the present invention, the conflict is eliminated according to a preset conflict resolution method, including:

[0072] Conflicts are resolved using similarity methods, semantic distance methods, or proximity methods.

[0073] Preferably, in the multidimensional data fault decision method provided in the embodiment of the present invention, a reverse branch reasoning method is selected to perform reasoning and generate a decision result, including:

[0074] Develop failure hypotheses;

[0075] Determine whether the prerequisites related to the validity of the hypothesis are known;

[0076] If known, the preset matching rule is used to match the fault feature fact with the rules in the preset fault rule base. If the match is successful, the confidence is calculated. If the calculated confidence is greater than or equal to the set threshold, a decision result is generated;

[0077] If it is unknown, the premise conditions are verified to find out all premise conditions related to the assumption, and all the premise conditions found become assumptions, and the step of judging whether the premise conditions related to the assumption is known is re-executed.

[0078] Preferably, in the above-mentioned multidimensional data fault decision method provided by the embodiment of the present invention, the confidence is calculated using the following formula:

[0079] T=∑ω i ×θ i (i=1,2,...,n)

[0080] Among them, ω i is the weight coefficient of the prerequisite, which represents the measure of the impact of the symptoms in the prerequisite on the fault, θ i is the confidence of the premise, which indicates the probability that the premise may lead to the conclusion.

[0081] An embodiment of the present invention further provides a multidimensional data fault decision device, comprising a processor and a memory, wherein the processor implements the multidimensional data fault decision method provided in the embodiment of the present invention when executing a computer program stored in the memory.

[0082] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program, wherein when the computer program is executed by a processor, the multidimensional data fault decision method provided by the embodiment of the present invention is implemented.

[0083] It can be seen from the above technical scheme that a multi-dimensional data fault decision method provided by the present invention includes: obtaining different monitoring data of the equipment and extracting characteristic values; performing fuzzy processing on the extracted characteristic values; performing fuzzy reasoning on the processed characteristic values ​​according to a preset fault rule library to generate a decision result; defuzzifying the generated decision result to obtain the final fault decision result and maintenance suggestion.

[0084] The present invention obtains the final fault decision result and maintenance suggestion by extracting characteristic values, fuzzifying the characteristic values, performing fuzzy reasoning and defuzzification using a rule base, which can make the reasoning process more reasonable, ensure the objectivity and correctness of the decision result, make full use of fault information, accurately identify the fault location and cause, and provide targeted maintenance decision suggestions in a timely manner.

[0085] In addition, the present invention also provides a corresponding device and a computer-readable storage medium for the multidimensional data fault decision method, which further makes the above method more practical. The device and the computer-readable storage medium have corresponding advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0087] Figure 1 A flowchart of a multidimensional data fault decision-making method provided by an embodiment of the present invention;

[0088] Figure 2 A schematic diagram of the overall architecture of fuzzy reasoning provided by an embodiment of the present invention;

[0089] Figure 3 A schematic diagram of the overall architecture of the multidimensional data fault decision-making method provided by an embodiment of the present invention;

[0090] Figure 4 One of the flow charts for establishing a preset fault rule base provided by an embodiment of the present invention;

[0091] Figure 5 The second flowchart of establishing a preset fault rule base provided by an embodiment of the present invention;

[0092] Figure 6 A flowchart of forward fuzzy reasoning provided by an embodiment of the present invention;

[0093] Figure 7 A flowchart of reverse branch reasoning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0094] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.

[0095] The present invention provides a multi-dimensional data fault decision method, such as Figure 1 and Figure 2 As shown, the following steps are included:

[0096] S101, obtaining different monitoring data of the equipment and extracting characteristic values;

[0097] In practical applications, the acquisition module can be used to obtain different monitoring data of the equipment, that is, to collect data from different monitoring data sources, and use certain means to integrate all relevant signals from different monitoring data sources to obtain multi-source signals. From the multi-source signals, relevant feature values ​​corresponding to different fault characteristics can be extracted, which is helpful for better subsequent fusion judgment.

[0098] S102, performing fuzzy processing on the extracted feature values;

[0099] S103, using a preset fault rule library to perform fuzzy reasoning on the processed feature values ​​to generate a decision result;

[0100] It should be noted that fuzzy reasoning is a thinking process that derives possible fault causes from one or several known judgments based on characteristic values ​​and using the fuzzy reasoning strategy set in the fuzzy fault rule library. Figure 2 It shows that fuzzy reasoning can be established and process fuzzy information based on fuzzy concepts and fuzzy logic.

[0101] S104: Defuzzify the generated decision result to obtain the final fault decision result and maintenance suggestion.

[0102] In the above-mentioned multidimensional data fault decision method provided by the embodiment of the present invention, the final fault decision result and maintenance suggestion are obtained by extracting characteristic values, fuzzifying the characteristic values, using the rule base for fuzzy reasoning and defuzzification. This can make the reasoning process more reasonable, ensure the objectivity and correctness of the decision result, make full use of fault information, accurately identify the fault location and cause of the fault, and provide targeted maintenance decision suggestions in a timely manner.

[0103] It should be noted that the following equipment takes bearings as an example, but the present invention is not limited to the field of bearing diagnosis, but is applicable to all mechanical equipment fault diagnosis fields.

[0104] In specific implementation, in the above multidimensional data fault decision method provided in the embodiment of the present invention, if Figure 3 As shown, the characteristic values ​​may include any of frequency domain characteristic values, abnormal sound characteristic values, time domain characteristic values, impact signal characteristic values, abrasive characteristic values, physical and chemical characteristic values, temperature characteristic values, pressure characteristic values, viscosity characteristic values ​​or other characteristic values ​​obtained by mechanism analysis.

[0105] It can be understood that mechanism analysis refers to a method that is different from statistics and machine learning. Specifically, characteristic values ​​can be obtained by analyzing equipment geometry, kinematics, tribology, equipment operation laws, equipment life degradation laws, etc.

[0106] According to the different monitoring data sources, it can be divided into noise, vibration signal, impact signal, lubricating oil and other data sources; when the data source is noise, the frequency domain eigenvalue and / or abnormal sound eigenvalue can be extracted; when the data source is a vibration signal, the time domain eigenvalue and / or frequency domain eigenvalue can be extracted; when the data source is an impact signal, the impact signal eigenvalue can be extracted; when the data source is lubricating oil, the abrasive eigenvalue and / or physical and chemical eigenvalue can be extracted; when the data source is other data sources, the temperature eigenvalue, pressure eigenvalue and / or viscosity eigenvalue can be extracted. The present invention can analyze the characteristics of different data source responses for different fault types. For example, for identifying the fault of shaft imbalance, the amplitude of the rotor vibration fundamental frequency in the vibration signal is the most important physical quantity, and for the fault of shaft misalignment, the amplitude of the rotor 2 times the fundamental frequency in the vibration signal is the most important physical quantity.

[0107] Taking bearings as an example, the following detailed descriptions are given on the generation mechanisms of bearing noise, bearing vibration, impact, lubricating oil, and other data sources, to facilitate the subsequent understanding of the implementation methods.

[0108] First, the mechanism of bearing noise generation is explained: the main reason for bearing noise is that when the bearing rotates, under the action of load, the bearing deforms and causes vibration and noise. Common noises include bearing ring noise, cage noise, rolling element falling noise, and friction noise. In addition, it is also related to the geometry, waviness, eccentricity, runout of the raceway, and the roundness and geometry of the rolling element of the bearing ring. When burrs, rust spots, pitting, peeling and other defects appear on the surface of the rolling bearing raceway or the rolling element, the bearing will produce stress changes during operation, causing high-frequency vibration. This vibration will propagate outward in the form of solid sound. If these defects appear on the raceway, the noise generated is often continuous; if they occur on the rolling element, the noise appears periodically from time to time. If impurities and granular dust fall into the rolling bearing, it may also cause it to emit unstable noise. The radial clearance of the bearing is also a factor that causes noise. If the clearance is too small, a high-pitched noise will be emitted, and if the clearance is too large, a loud roar will be generated.

[0109] Bearing noise is caused by many factors that produce bearing vibration, among which the surface roughness and waviness of the ring raceway and rolling elements have the greatest impact.

[0110] Important sources of bearing noise also include the impact sound between the rolling elements and the cage, the whistling sound caused by the vortex of the cage, the contact and friction sound between the rolling elements and the raceway (when the lubrication condition is poor), etc.

[0111] The background noise of the bearing sound frequency has the characteristics of white noise, but the main frequency components of the abnormal noise are concentrated in 1000Hz~10,000Hz, that is, the bearing noise generally manifests as medium and high frequency noise.

[0112] Noise detection can be used to diagnose rolling bearing faults by monitoring the elastic waves emitted by the bearing during the process of crack initiation, expansion and damage. When bearings are subjected to alternating loads or impact forces, elastic waves will be generated by the crystal dislocation, grain boundary displacement and crack generation and expansion of metal materials. There will be differences in the noise signals of different types of bearing faults. By identifying these differences, rolling bearing faults can be detected and located.

[0113] Second, explain the mechanism of bearing vibration: the vibration signal contains a wealth of mechanical state information. The mechanical design, installation quality and whether the operation is normal can all be reflected in the vibration signal. After the vibration signal is converted into an electrical signal, it can be processed into a variety of characteristic information spectra that can reflect the fault state, providing a basis for fault identification. At present, in the process of monitoring the operating status of the bearing, amplitude parameters, characteristic frequencies, autoregressive parameters, fractal dimensions, cyclostationarity, etc. are selected as characteristic parameters. There are two main forms of vibration generated by rolling bearings during operation:

[0114] The first is the vibration related to the elasticity of the bearing itself. This vibration is a characteristic of the bearing itself, and it is impossible to establish a connection with the fault characteristics from the signal of this type of vibration;

[0115] The second is abnormal vibration caused by minor damage to the bearing surface. When the rolling element of the rolling bearing is damaged, an alternating periodic exciting force will appear. Due to the various forms of damage, the vibration caused by this exciting force will contain different frequency components. The operating state and damage state of the bearing jointly determine the fault characteristic frequency, including the basic frequency of the bearing, the passing frequency of the bearing, and the natural frequency of the bearing fault.

[0116] When a bearing has any fault, its vibration signal will definitely change. From the perspective of vibration mechanism, the frequency components and frequency distribution in the signal will be significantly different from the normal frequency, which is called the fault characteristic frequency.

[0117] Vibration sensors obtain the vibration acceleration signals of bearings during operation, and evaluate the bearing status based on the analysis of the measured signals. It has the advantages of easy signal monitoring, simple processing, sensitivity to early bearing failures, and easy applicability.

[0118] Third, the mechanism of impact generation is explained: the low-frequency impact force generated when the bearing fails will arouse the high-frequency resonance response of the bearing element or sensor. The resonance is in the high-frequency band and is not easily disturbed by low-frequency mechanical faults (such as misalignment and imbalance). The sensor is sensitive to vibration signals in the high-frequency band. The signal is narrow-band filtered and envelope detected. The amplitude of the obtained signal reflects the magnitude of the impact force, that is, the magnitude of the rolling bearing fault. Envelope analysis mainly includes two processes: resonance signal picking and envelope detection. That is, picking up the signal near the resonance peak (that is, the natural frequency of the sensor or mechanical system) or at the resonant frequency of the high-frequency resonator to obtain a high-frequency vibration signal with a high signal-to-noise ratio, and then obtaining an envelope waveform through the envelope detection method. The envelope waveform contains characteristic information related to rolling bearing fault diagnosis, which includes the outer ring of the cage, the inner ring of the cage, the outer ring, the inner ring, the roller end face, the roller circumference, the gear and the adjacent gear.

[0119] Fourth, explain the mechanism of lubricating oil production: Engine failures mainly occur in the bearings, gears and other parts of the engine transmission system and accessory system. The wear and tear produced in the lubricating oil includes fatigue wear, abrasive wear, adhesive wear, corrosive wear and erosion wear.

[0120] Fatigue wear is when two contact surfaces make rolling or rolling-sliding composite relative motions, and due to the action of alternating stress, the contact surface metal undergoes plastic deformation, cracks are generated, and after the cracks expand, the material breaks and peels off, leaving a pit on the metal surface. Contact fatigue has no fatigue limit, and the fatigue life fluctuates greatly. Generally, there are two types of fatigue wear: convergent and expansive. Convergent fatigue wear often occurs during the running-in period, and generally disappears on its own with the end of running-in; expansive fatigue wear is that the initial fatigue pits continue to expand with the growth of time until the parts fail.

[0121] Abrasive wear is the material loss or transfer caused by a surface contacting a hard micro-convex body on its matching surface, or contacting a hard particle that moves relative to a pair of contact surfaces. The former is called two-body abrasive wear, and the latter is called three-body abrasive wear. Engine parts are more prone to two-body abrasive wear during the running-in period. There are three damage principles according to the different conditions of abrasive grinding on the metal surface: micro-cutting, in this damage process, due to the geometric shape of the abrasive, complete chips can be produced, and the surface damage may be furrows or flash. For brittle materials, micro-cutting will cause material collapse; fatigue failure, due to the action of abrasive particles, the metal surface produces alternating contact stress and causes fatigue failure, and the ridges of the furrows will also cause multiple strains in the metal, resulting in fatigue failure; indentation For metal surfaces with greater plasticity, abrasive particles press into the metal surface under the action of force to produce indentations, and squeeze out peeling from the surface.

[0122] Adhesive wear is the contact of micro-convex bodies on the friction surface of mechanical parts. Due to excessive stress, local plastic deformation occurs. The plastic surface adheres under the action of load to form node welds. Under the condition of relative motion, the node material may be transferred from one softer surface to another, or it may be torn off to form adhesive wear abrasive particles. According to the degree of surface damage, adhesive wear is divided into five levels, from light to heavy: slight wear, smearing, abrasion, tearing and biting. The bonding of engine gears, bearings and other parts under high speed, high temperature and heavy load is a serious form of adhesive wear damage.

[0123] Corrosion and wear is the loss of surface material caused by chemical or electrochemical reactions between the metal surface and the surrounding medium. When flying over the sea or using inferior fuel, engine airflow channel parts are prone to corrosion and wear.

[0124] Erosion wear is the loss of surface material caused by the erosion of the metal surface by fluid (liquid or gas) containing hard particles. According to the incident angle of the particles, erosion wear can be divided into grinding erosion (incident angle close to 0°) and collision erosion (incident angle close to 90°). For example, when a helicopter works in dusty conditions, the engine airflow channel parts are prone to erosion wear.

[0125] Wear products refer to tiny particles produced during the wear process, so they are also called wear particles, or abrasive particles for short. Abrasive particles are the most direct information element for revealing the wear mechanism of tribological surfaces, monitoring the wear process, and diagnosing the type of wear failure. Because abrasive particles are affected by different contact modes, loads, temperatures, and environmental factors during their formation, and have different wear mechanisms, abrasive particles have different morphological characteristics. According to the morphological characteristics of abrasive particles, they can be divided into: normal sliding abrasive particles, severe sliding abrasive particles, cutting abrasive particles, fatigue peeling abrasive particles, layered abrasive particles, spherical abrasive particles, and oxide abrasive particles.

[0126] Online monitoring of lubricating oil is mainly achieved through online lubricating oil sensors. It is very effective when the bearing surface has wear debris larger than 200μm, but it is difficult to locate the fault.

[0127] Fifth, give examples of other data sources: For example, temperature is a basic operating parameter during the operation of rolling bearings, and its magnitude is mainly affected by the radial / axial load of the bearing, the relative speed of the inner and outer rings, the lubrication conditions and the bearing status. In the case of poor lubrication, temperature monitoring is more sensitive. When the temperature reacts, it is already in a more serious fault.

[0128] According to the above analysis, based on the different fault characteristics that different data sources can reflect, relevant feature values ​​that can identify the characteristics can be selected.

[0129] In specific implementation, in the above-mentioned multidimensional data fault decision method provided by an embodiment of the present invention, when executing step S101, when the characteristic value is a time domain characteristic value, the method of extracting the time domain characteristic value may specifically include: selecting time domain statistical parameters that are more sensitive to the signal as time domain characteristic values, such as peak-to-peak value, mean value, root mean square value, standard deviation, skewness coefficient, kurtosis, amplitude of the rotor vibration fundamental frequency, amplitude of the rotor vibration twice the fundamental frequency, etc., that is to say, any of the peak-to-peak value, mean value, root mean square value, standard deviation, skewness coefficient, kurtosis, amplitude of the rotor vibration fundamental frequency, amplitude of the rotor vibration twice the fundamental frequency can be extracted as the time domain characteristic value.

[0130] In specific implementation, in the above-mentioned multidimensional data fault decision method provided by an embodiment of the present invention, when executing step S101, when the eigenvalue is a frequency domain eigenvalue, the method of extracting the frequency domain eigenvalue may specifically include: using a fast intrinsic mode decomposition algorithm (FIMD) algorithm to decompose the original signal to obtain an intrinsic mode FIMD component; reconstructing the intrinsic mode FIMD component to obtain a reconstructed signal; using Fourier transform to extract frequency domain features from the time domain information of the reconstructed signal to obtain a signal change trend in the frequency domain, and extracting information on the frequency band energy, frequency band width, and whether there is an abnormal frequency.

[0131] Specifically, firstly, the original signal is adaptively decomposed based on FIMD to eliminate the modal aliasing phenomenon:

[0132] Assume that the input original signal is f(t), and calculate the extreme point R(t) of the original signal f(t):

[0133] R(t), t0≤t j ≤t m-1 (1)

[0134] Among them, t0, t j and t m-1 They represent the sampling moments of different extreme points respectively.

[0135] The extreme points are transformed using a linear transformation method and the corresponding residual r(t) is generated.

[0136]

[0137] Among them, t j+1 is the sampling time corresponding to the j+1 extreme points.

[0138] Subtract the residual from the original signal to obtain the inherent modal function of the original signal. The implementation process is as follows:

[0139] The i-th extreme point of the intrinsic mode function E(t i) as the control point of the correction, and define the coordinates of the control point as P i , whose expression is as follows:

[0140] P i (t i ,E(t i )), t0≤t i ≤t m-1 (3)

[0141] Among them, t i is the sampling time corresponding to the extreme point i.

[0142] The product of the i-th vector is defined as follows:

[0143] α i =P i-1 P i ×P i P i+1 (4)

[0144] Therefore, the median of the control point can be calculated based on the product of two adjacent vectors. At the same time, the median of the control point is connected by multiple (such as cubic) spline connection method to correct the residual. At this time, the corrected residual is subtracted from the original input signal to obtain the corrected natural mode component.

[0145] The process is iterated continuously, and the decomposition calculation is stopped when the number of extreme value points of the residual is less than 3.

[0146] Therefore, the original input signal can be decomposed into multiple unique intrinsic mode FIMD components and a residual component. Then the original signal can be expressed as:

[0147]

[0148] Then, the decomposed FIMD components are reconstructed to obtain the denoised signal. This process may include:

[0149] Calculate the correlation degree of each order FIMD component The formula is as follows:

[0150]

[0151] in, is the cross-correlation function between the FIMD component and the original signal, R S (t) is the autocorrelation function of the original signal.

[0152] choose The larger FIMD component is reconstructed to obtain the reconstructed signal f′(t).

[0153] Finally, Fourier transform is used to extract frequency domain features from the signal's time domain information, obtain and view the signal change trend in the frequency domain, and extract information such as frequency band energy, frequency band width, and whether there is abnormal frequency in the results based on the fault mechanism.

[0154] In specific implementation, in the above-mentioned multidimensional data fault decision method provided by an embodiment of the present invention, when executing step S101, when the characteristic value is the characteristic value of the impact signal, the method of extracting the characteristic value of the impact signal may specifically include: using a resonator to absorb low-frequency fault impacts, generating high-frequency resonance to obtain a free decay oscillation waveform, and filtering out low-frequency vibration signals; performing envelope demodulation on the resonance waveform, and performing low-pass filtering on the obtained envelope signal; performing Fourier transform on the envelope signal after low-pass filtering to obtain an envelope spectrum; according to the envelope spectrum, extracting any kind of characteristic values ​​of the retainer to the outer ring, the retainer to the inner ring, the outer ring, the inner ring, the roller end face, the roller circumferential surface, the current gear, and the adjacent gear as the impact signal characteristic value, and the characteristic value is the corresponding amplitude or dB value on the envelope spectrum.

[0155] Specifically, first, a resonator is used to absorb low-frequency fault shocks, generate high-frequency resonance (generalized resonance) to obtain a free decay oscillation waveform, filter out unnecessary low-frequency vibration signals, and then perform envelope demodulation on the resonance waveform, and perform low-pass filtering on the envelope signal and perform Fourier transform to obtain the envelope spectrum.

[0156] Among them, Hilbert transform can be used to perform envelope demodulation on the resonance waveform, and the formula is as follows:

[0157]

[0158] in, represents the envelope signal obtained after Hilbert transform, x(r) represents the input resonance waveform signal, t and r both represent time, corresponding to the time series.

[0159] In addition, the envelope demodulation of the resonance waveform can also be performed by using absolute value detection to obtain an envelope signal.

[0160] Since the fault will cause a large peak at the fault characteristic frequency and its multiples in the envelope spectrum, in the specific implementation, in the above multidimensional data fault decision method provided by the embodiment of the present invention, while extracting the corresponding amplitude or dB value on the envelope spectrum, it can also include: selecting the energy proportion of the corresponding characteristic frequency and its multiples in the total energy of the entire envelope spectrum as the impulse signal characteristic value, the formula is as follows:

[0161]

[0162] in, Indicates the obtained proportion value, Represents the energy sum of the fault feature and its multiples; Represents the energy and frequency of the entire envelope spectrum; y i and j Represents the amplitude at each frequency.

[0163] In specific implementation, in the above multidimensional data fault decision method provided by the embodiment of the present invention, when executing step S101, when the characteristic value is the abrasive characteristic value, the method of extracting the abrasive characteristic value may specifically include: identifying ferromagnetic and non-ferromagnetic particles through online monitoring of lubricating oil abrasive particles, and at the same time, the number and quality of particles in each size range can be counted. In practical applications, other features can be extracted by monitoring moisture, viscosity, temperature, and pressure through different sensors, and directly extracting related parameters.

[0164] In specific implementation, in the above-mentioned multidimensional data fault decision method provided by the embodiment of the present invention, step S102 performs fuzzy processing on the extracted eigenvalues, which may specifically include: fuzzy processing on the eigenvalues ​​through a membership function. In practical applications, a peeling test of the entire process of running to failure can be performed to modify the alarm threshold of the fuzzy set and select a suitable membership function. The fuzzy set may include: A: normal, B: warning, C: first-level alarm, D: second-level alarm. The eigenvalues ​​can be converted into fuzzy quantities that can be processed by the fusion fault diagnosis method based on fuzzy logic reasoning, namely, the membership function, through the membership function. Ways to determine the membership function may include: expert determination method, borrowing existing objective scales, fuzzy statistics, comparative sorting method, comprehensive weighting method, basic concept expansion method, etc.

[0165] In specific implementation, in the above-mentioned multidimensional data fault decision method provided in an embodiment of the present invention, before executing step S103 to perform fuzzy reasoning on the processed feature value according to the preset fault rule base, it can also include: establishing relevant rules in the preset fault rule base and performing rule verification; if there is no rule conflict, storing the rule in the preset fault rule base; if there is a rule conflict, modifying the rule or the original rule according to the corresponding knowledge of the fault mechanism, and storing the rule in the preset fault rule base when there is no rule conflict.

[0166] Figure 4 and Figure 5 The specific steps of establishing the preset fault rule base are shown. Figure 4 As shown in the figure, the new rule input step is to input the rules in the prescribed format according to the analysis of test data, expert experience and engineering knowledge, and perform rule verification. When there is no rule conflict, it is stored in the preset fault rule library. When there is a rule conflict, a prompt is given and the input rule is modified. Figure 5As shown, the rule modification step is to find out that the original rule is not applicable through actual data, search for the original rule in the rule base and modify it, and then store the modified rule in the rule base.

[0167] In addition, it should be noted that in fault decision-making, due to the fuzziness of fault characteristics and the uncertainty of diagnostic experience knowledge, in order to enable fault decision-making to make correct judgments under incomplete information, the rules are expressed using fuzzy production rules that combine production rules and fuzzy mathematics. Fuzzy production rules have the advantages of effectively expressing heuristic knowledge, and can give a credibility factor based on data reliability, thereby realizing fuzzy reasoning.

[0168] Therefore, in a specific implementation, in the above-mentioned multidimensional data fault decision method provided in an embodiment of the present invention, while establishing relevant rules in the preset fault rule base in the above-mentioned steps, it can also include: setting the weight coefficient of the premise condition, the rule activation threshold and the rule credibility for the preset fault rule base to make the rule fuzzy, that is, the rule is a fuzzy production rule.

[0169] It should be noted that the weight coefficients of the relevant rules or preconditions in the above-mentioned preset fault rule base can be established based on mechanism analysis, that is, the establishment of the preset fault rule base makes full use of the analysis of the equipment operation and signal generation mechanism, and establishes the weights of relevant rules or each characteristic value based on mechanism analysis. However, the existing technology generally relies on a large number of cases, or randomly sets some rules and weights in advance, so that the accuracy of the later model is low, and a lot of learning is required to obtain better results.

[0170] Specifically, the fuzzification of rules in the fault rule base is mainly achieved by setting the weight coefficient of the prerequisite, setting the rule activation threshold, and setting the rule credibility. The specific expressions are as follows:

[0171]

[0172] The above formula is a fuzzy relation expression, where R i represents the i-th inference rule, A i1 , A i2 , …, A in represents the premise of the rule, ω i1 ,ω i2 ,…,ω in is the weight coefficient of the corresponding prerequisite, B i represents the conclusion of the rule, CF represents the credibility of the rule, and τ represents the activation threshold of the rule. The credibility of each rule is different. Adding a credibility factor to the rule indicates the degree of confidence of the human expert in the rule, thus adding the ability of fuzzy representation on the basis of traditional production rules.

[0173] τ is the activation threshold of the rule (the limit that can be used). Fuzzy matching is performed on the fuzzy signs in the fault decision and the premise conditions of the fuzzy production rules in the fault rule base. The matching degree obtained by the present invention through fuzzy matching is k (0≤k≤1). When k>τ, the rule is used; if τ=1, the rule is an accurate rule. ω1、ω2、…、ω n is the weight coefficient of the premise condition, which indicates the measure of the impact of the symptom in the premise sub-condition on the fault. Different weight coefficients are assigned to different symptoms in the same fault; different weight coefficients should be assigned to different faults for the same symptom. The higher the credibility of the same symptom to a fault, the greater the weight coefficient, where the sum of all weight coefficients is 1. The weight threshold is set according to different types and different working conditions of the actual engine.

[0174] In specific implementation, in the above-mentioned multidimensional data fault decision method provided by the embodiment of the present invention, in order to reduce the steps of reasoning and ensure the efficiency of reasoning, the reasoning mechanism can adopt a hybrid reasoning method. Step S103 performs fuzzy reasoning on the processed eigenvalues ​​according to the preset fault rule library to generate a decision result. Specifically, it can include: performing forward fuzzy reasoning on the processed eigenvalues ​​according to the preset fault rule library to generate a decision result; if the forward fuzzy reasoning fails and no decision result is generated, select the reverse branch reasoning method to perform reasoning and generate a decision result. In other words, if there is no conclusion required for reasoning during the forward reasoning process, the required conclusion is solved with the help of reverse reasoning, so that the data fault decision method has a strong pertinence. In this way, the combination of forward and reverse reasoning improves the efficiency of reasoning, can provide targeted maintenance decision suggestions in a timely manner, ensures the real-time and accuracy of diagnosis, and has good engineering practicality.

[0175] Furthermore, in a specific implementation, in the above-mentioned multidimensional data fault decision method provided in an embodiment of the present invention, the above-mentioned steps perform forward fuzzy reasoning on the processed feature values ​​according to the rule base, which may specifically include: storing the processed feature values ​​as fault feature facts in the fact base; using preset matching rules (such as Rete matching algorithm) in the preset fault rule base to match the fault feature facts with the rules in the preset fault rule base; if the match is successful, calculating the confidence, and if the calculated confidence is greater than or equal to the set threshold, generating a decision result; if the calculated confidence is less than the set threshold, determining that the forward fuzzy reasoning has failed.

[0176] In the present invention, in order to improve the accuracy of the confidence, after the fault feature fact is successfully matched with the rules in the preset fault rule library, before calculating the confidence, it can also include: when there are multiple rules that are matched at the same time, the conflict is eliminated according to the preset conflict resolution method. When there is no conflict, the confidence is calculated, so that a more accurate confidence can be obtained.

[0177] Specifically, Figure 6 As shown, the detailed process of forward fuzzy reasoning for fault decision-making can include the following steps:

[0178] Step 1: According to the previous representation method of fault feature facts, store the fault feature facts into the fact database.

[0179] Step 2: Use preset matching rules (such as Rete matching algorithm) in the preset fault rule library to match the rules in the system, and determine whether there are applicable rules in the preset fault rule library. If the match is successful, when there are multiple rules that match at the same time, jump to step 3; otherwise jump to step 4.

[0180] Step 3: Eliminate the conflict according to the preset conflict resolution strategy, otherwise jump to step 5.

[0181] Commonly used conflict resolution methods include similarity method, semantic distance method and closeness method, for example, the Hamming progress calculation method in the fuzzy progress method is used:

[0182]

[0183] Among them, A is the fuzzy alarm and B is the antecedent of the fuzzy key rule.

[0184] Step 4: Check whether there is a solution in the working database. If so, exit and generate a decision result, otherwise enter the reverse branch reasoning.

[0185] Step 5: Calculate the confidence level of the conclusion using the following formula:

[0186] T=∑ω i ×θ i (i=1,2,…,n) (11)

[0187] Among them, ω i Prerequisite A i The weight coefficient of the prerequisite A i The measure of the impact of the symptom on the fault, θ i Prerequisite A i The confidence level indicates that the premise A i The probability that the conclusion may occur.

[0188] If the calculated confidence is greater than or equal to the set threshold, the decision result is generated, the cause of the fault, the maintenance plan and the suggestion are obtained, and the decision is completed. Otherwise, jump to step 2. If the calculated confidence is less than the set threshold, it is determined that the forward fuzzy reasoning fails and turn to the reverse branch reasoning.

[0189] It should be noted that the Rete matching algorithm is a fast forward rule matching algorithm and an efficient pattern matching algorithm for production systems. Its matching speed is independent of the number of rules.

[0190] Furthermore, in specific implementation, in the above-mentioned multidimensional data fault decision method provided by an embodiment of the present invention, the reverse branch reasoning method is selected in the above steps to perform reasoning and generate a decision result, which may specifically include: establishing a fault hypothesis; judging whether the premise conditions related to the establishment of the hypothesis are known; if known, a preset matching rule (such as a Rete matching algorithm) is used to match the fault feature facts with the rules in a preset fault rule library. If the match is successful, the confidence is calculated. If the calculated confidence is greater than or equal to the set threshold, a decision result is generated; if unknown, the premise conditions are verified to find out all the premise conditions related to the establishment of the hypothesis, and all the premise conditions found become hypotheses, and the step of judging whether the premise conditions related to the establishment of the hypothesis are known is re-executed.

[0191] Specifically, Figure 7 As shown, the specific process of the reverse branch push mode may include the following steps:

[0192] Step 6: First establish a fault hypothesis: assume that the fault has occurred.

[0193] Step 7: Check the rule antecedents: Determine whether the premise conditions related to the establishment of the hypothesis are known. If the rule antecedent conditions are known, jump to step 8, otherwise jump to step 9.

[0194] Step 8. Match the known facts with the rules. If the match is successful, calculate the confidence of the conclusion according to the formula. If the calculated confidence is greater than or equal to the set threshold, search the decision conclusion table, output the cause of the fault and the maintenance plan, and the decision is successful. If the calculated confidence is less than the set threshold, determine whether there are any assumptions. If there are any assumptions, jump to step 7; otherwise, jump out of this decision and the decision fails.

[0195] Step 9: If the premise condition is unknown (which means that the premise condition is an intermediate condition), it is necessary to verify the premise condition. Then, all premise conditions for the assumption to be true will be found, and all premise conditions will become assumptions, and jump to step 7.

[0196] In specific implementation, in the above-mentioned multidimensional data fault decision method provided in an embodiment of the present invention, step S104 defuzzifies the generated decision results, which may specifically include: since the decision results of the present invention are: normal, warning, first-level alarm, and second-level alarm, the membership distribution of the fuzzy alarm on this group of language variables indicates the overall situation of the fuzzy alarm, and the fuzzy alarm can be defuzzified by processing the membership of the fuzzy alarm on each language variable, thereby providing the user with a clear and easy-to-understand conclusion. The defuzzification method may adopt the "center of gravity" method, the maximum membership method, and the like.

[0197] It can be understood that in the above-mentioned multi-dimensional data fault decision method provided by the present invention, data features of different data sources are selected according to the fault mechanism, so that the state of the same fault can be reflected from the multi-source data features, and the fault information is fully utilized to accurately locate the fault position and find out the cause of the fault; in addition, a preset fault rule library is established according to the fault mechanism, so that the reasoning process is more reasonable, and finally the credibility of the occurrence of the fault is given, ensuring the objectivity and correctness of the decision result; and, the fault decision fuzzy reasoning method uses a combination of positive and negative reasoning to improve the efficiency of reasoning, and can provide targeted maintenance decision suggestions in a timely manner, ensuring the real-time and accuracy of the diagnosis, and has good engineering practicality.

[0198] Based on the same inventive concept, an embodiment of the present invention also provides a multidimensional data fault decision device. Since the principle of solving the problem by the device is similar to that of the aforementioned multidimensional data fault decision method, the implementation of the device can refer to the implementation of the multidimensional data fault decision method, and the repeated parts will not be repeated.

[0199] In specific implementation, the multidimensional data fault decision-making device provided by the embodiment of the present invention may specifically include:

[0200] The feature value extraction module is used to obtain different monitoring data of the equipment and extract feature values;

[0201] A fuzzification module is used to perform fuzzification processing on the extracted feature values;

[0202] The fuzzy reasoning module is used to perform fuzzy reasoning on the processed feature values ​​according to the preset fault rule library to generate a decision result;

[0203] The defuzzification module is used to defuzzify the generated decision results to obtain the final fault decision results and maintenance suggestions.

[0204] In the above-mentioned multidimensional data fault decision-making device provided in the embodiment of the present invention, the interaction of the above-mentioned four modules can make the reasoning process more reasonable, ensure the objectivity and correctness of the decision result, make full use of fault information, accurately identify the fault location and cause of the fault, and provide targeted maintenance decision suggestions in a timely manner.

[0205] For more specific working processes of the above modules, please refer to the corresponding contents disclosed in the above embodiments, which will not be repeated here.

[0206] Correspondingly, an embodiment of the present invention further discloses a multidimensional data fault decision device, comprising a processor and a memory; wherein the processor implements the multidimensional data fault decision method disclosed in the above embodiment when executing a computer program stored in the memory.

[0207] For more specific processes of the above method, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0208] Furthermore, the present invention also discloses a computer-readable storage medium for storing a computer program; when the computer program is executed by a processor, the multi-dimensional data fault decision method disclosed above is implemented.

[0209] For more specific processes of the above method, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0210] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device and storage medium disclosed in the embodiment, since they correspond to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0211] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0212] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0213] In summary, a multidimensional data fault decision method provided by an embodiment of the present invention includes: obtaining different monitoring data of the equipment and extracting characteristic values; performing fuzzification processing on the extracted characteristic values; performing fuzzy reasoning on the processed characteristic values ​​according to a preset fault rule base to generate a decision result; defuzzifying the generated decision result to obtain the final fault decision result and maintenance suggestion. The present invention obtains the final fault decision result and maintenance suggestion by extracting characteristic values, performing fuzzification processing on the characteristic values, performing fuzzy reasoning using a rule base, and defuzzification. This can make the reasoning process more reasonable, ensure the objectivity and correctness of the decision result, make full use of fault information, accurately identify the fault location and fault cause, and provide targeted maintenance decision suggestions in a timely manner. In addition, the present invention also provides a corresponding device and a computer-readable storage medium for the multidimensional data fault decision method, which further makes the above method more practical, and the device and computer-readable storage medium have corresponding advantages.

[0214] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0215] The multidimensional data fault decision method, device and storage medium provided by the present invention are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A multidimensional data fault decision method, characterized in that: include: Obtain different monitoring data of the equipment and extract feature values; When the eigenvalue is a frequency domain eigenvalue, extracting the frequency domain eigenvalue includes: calculating the extreme value points of the original signal; transforming the extreme value points of the original signal using a linear transformation method and generating corresponding residuals; transforming the i-th extreme value point E(t i ) as the modified control point P i , where t i is the sampling time corresponding to the extreme point i; the control point P is calculated according to the product of two adjacent vectors i The median value of α i =P i-1 P i ×P i P i+1 , where P i-1 and P i+1 is the control point P i Adjacent vectors; At the same time, the control point P is connected by multiple spline connections i The median of is connected to correct the residual; the residual is subtracted from the original signal to obtain the intrinsic modal FIMD component corresponding to the original signal; the correlation degree of each order FIMD component is calculated using the following formula: in, It indicates the correlation degree of each order FIMD component; represents the cross-correlation function between the FIMD component and the original signal, R S (t) represents the autocorrelation function of the original signal; Select the FIMD component corresponding to the correlation degree greater than the set value for reconstruction to obtain a reconstructed signal; use Fourier transform to extract frequency domain features of the time domain information of the reconstructed signal to obtain the signal change trend in the frequency domain, and extract the frequency band energy, frequency band width and whether there is an abnormal frequency; Performing fuzzy processing on the extracted feature values; Performing forward fuzzy reasoning on the processed characteristic values ​​according to the preset fault rule library to generate a decision result; if the forward fuzzy reasoning fails, selecting a reverse branch reasoning method to perform reasoning to generate a decision result; wherein fuzzy reasoning is a thinking process of deriving possible fault causes from one or several known judgments based on the processed characteristic values ​​and using the fuzzy reasoning strategy set in the preset fault rule library; The generated decision result is defuzzified to obtain the final fault decision result and maintenance suggestion.

2. The multidimensional data fault decision method according to claim 1, characterized in that: The characteristic values ​​include any of frequency domain characteristic values, abnormal sound characteristic values, time domain characteristic values, impact signal characteristic values, abrasive characteristic values, physical and chemical characteristic values, temperature characteristic values, pressure characteristic values, viscosity characteristic values ​​or other characteristic values ​​obtained by mechanism analysis.

3. The multidimensional data fault decision method according to claim 1, characterized in that: When the eigenvalue is a time domain eigenvalue, extracting the time domain eigenvalue includes: Extract any of the peak-to-peak value, mean value, RMS value, standard deviation, skewness coefficient, kurtosis, amplitude of the rotor vibration fundamental frequency, and amplitude of twice the rotor vibration fundamental frequency.

4. The multidimensional data fault decision method according to claim 1, characterized in that: The corresponding residual volume is generated using the following formula: Where r(t) represents the residual amount, t0, t j and t m-1 Respectively represent the sampling time of different extreme points of the original signal, t j+1 represents the sampling time corresponding to the j+1 extreme point, R(t j ) represents the sampling time t j The extreme point of R(t j ) represents the sampling time t j+1 The extreme point of .

5. The multidimensional data fault decision method according to claim 1, characterized in that: When the characteristic value is an impact signal characteristic value, extracting the impact signal characteristic value includes: Use the resonator to absorb low-frequency fault shocks, generate high-frequency resonance to obtain a free-decay oscillation waveform, and filter out low-frequency vibration signals; Performing envelope demodulation on the resonance waveform and performing low-pass filtering on the obtained envelope signal; Performing Fourier transform on the envelope signal after low-pass filtering to obtain an envelope spectrum; According to the envelope spectrum, any characteristic values ​​of the retainer to the outer ring, the retainer to the inner ring, the outer ring, the inner ring, the roller end face, the roller circumferential surface, the current gear, and the adjacent gear are extracted, and the characteristic value is the corresponding amplitude or dB value on the envelope spectrum.

6. The multidimensional data fault decision method according to claim 5, characterized in that: Envelope demodulation of the resonant waveform, including: The envelope signal is obtained by envelope demodulating the resonance waveform using Hilbert transform or absolute value detection.

7. The multidimensional data fault decision method according to claim 5, characterized in that: While extracting the corresponding amplitude or dB value on the envelope spectrum, it also includes: The proportion of the energy of the corresponding characteristic frequency and its multiple frequency in the total energy of the entire envelope spectrum is selected as the impulse signal characteristic value.

8. The multidimensional data fault decision method according to claim 7, characterized in that: The following formula is used to obtain the proportion of the energy of the corresponding characteristic frequency and its multiple frequency in the total energy of the entire envelope spectrum: in, Indicates the obtained proportion value, Represents the energy sum of the characteristic frequency and its multiples; Represents the energy and frequency of the entire envelope spectrum; y i and j Represents the amplitude at each frequency.

9. The multidimensional data fault decision method according to claim 1, characterized in that: When the characteristic value is an abrasive particle characteristic value, extracting the abrasive particle characteristic value comprises: Identify ferromagnetic and non-ferromagnetic particles through online monitoring of lubricating oil wear particles; At the same time, the number and mass of particles within each size range are counted as the abrasive particle characteristic values.

10. The multidimensional data fault decision method according to claim 1, characterized in that: The extracted feature value is subjected to fuzzy processing, comprising: The eigenvalues ​​are fuzzified by using a membership function.

11. The multidimensional data fault decision method according to claim 1, characterized in that: Before performing fuzzy reasoning on the processed characteristic value according to the preset fault rule base, the method further includes: Establish relevant rules in the preset fault rule base and perform rule verification; If no rule conflict occurs, the rule is stored in the preset fault rule library; If there is a rule conflict, the rule or the original rule is modified according to the corresponding knowledge of the fault mechanism, and when there is no rule conflict, the rule is stored in the preset fault rule library.

12. The multidimensional data fault decision method according to claim 11, characterized in that: When establishing the relevant rules in the preset fault rule library, it also includes: The preset fault rule base is provided with a weight coefficient of a premise condition, a rule activation threshold and a rule credibility to make the rules fuzzy.

13. The multidimensional data fault decision method according to claim 12, characterized in that: The weight coefficients of the relevant rules or premise conditions in the preset fault rule base are established based on mechanism analysis.

14. The multidimensional data fault decision method according to claim 13, characterized in that: The following formula is used to set the weight coefficient of the premise condition, the rule activation threshold and the rule credibility for the preset fault rule base: The above formula is a fuzzy relation expression, where R i represents the i-th inference rule, A i1 , A i2 , …, A in represents the premise of the rule, ω i1 ,ω i2 ,…,ω in is the weight coefficient of the corresponding prerequisite, B i represents the conclusion of the rule, CF represents the credibility of the rule, and τ represents the activation threshold of the rule.

15. The multidimensional data fault decision method according to claim 14, characterized in that: Performing forward fuzzy reasoning on the processed characteristic value according to a preset fault rule base includes: storing the processed characteristic value as a fault characteristic fact in a fact database; Using a preset matching rule in the preset fault rule base to match the fault feature fact with the rules in the preset fault rule base; If the match is successful, the confidence is calculated. If the calculated confidence is greater than or equal to the set threshold, a decision result is generated; if the calculated confidence is less than the set threshold, the forward fuzzy reasoning is judged to have failed.

16. The multidimensional data fault decision method according to claim 15, characterized in that: After a successful match, but before calculating the confidence, it also includes: When multiple rules are matched at the same time, the conflicts are resolved according to the preset conflict resolution method.

17. The multidimensional data fault decision method according to claim 16, characterized in that: Eliminate conflicts according to the preset conflict resolution methods, including: Conflicts are resolved using similarity methods, semantic distance methods, or proximity methods.

18. The multidimensional data fault decision method according to claim 17, characterized in that: Select the reverse branch reasoning method to perform reasoning and generate decision results, including: Develop failure hypotheses; Determine whether the prerequisites related to the validity of the hypothesis are known; If known, the preset matching rule is used to match the fault feature fact with the rules in the preset fault rule base. If the match is successful, the confidence is calculated. If the calculated confidence is greater than or equal to the set threshold, a decision result is generated; If it is unknown, the premise conditions are verified to find out all premise conditions related to the assumption, and all the premise conditions found become assumptions, and the step of judging whether the premise conditions related to the assumption is known is re-executed.

19. The multidimensional data fault decision method according to claim 15 or 18, characterized in that: The confidence level is calculated using the following formula: T=∑ω i ×θ i (i=1,2,...,n) Among them, ω i is the weight coefficient of the prerequisite, which represents the measure of the impact of the symptoms in the prerequisite on the fault, θ i is the confidence of the premise, which indicates the probability that the premise may lead to the conclusion.

20. A multidimensional data fault decision-making device, characterized in that: It comprises a processor and a memory, wherein when the processor executes the computer program stored in the memory, the multidimensional data fault decision method as claimed in any one of claims 1 to 19 is implemented.

21. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein when the computer program is executed by a processor, the multidimensional data fault decision method according to any one of claims 1 to 19 is implemented.

Images