Motor fault diagnosis method based on inherent characteristic frequency

Abstract

The invention discloses a motor fault diagnosis method based on inherent characteristic frequency. A diagnosis system is constructed based on a conventional electromechanical fault simulation platform and signal acquisition equipment. The method comprises the steps that a multi-working-condition normal data feature library containing environment compensation factors is constructed, bidirectional acceleration signals are collected in real time through filtering, calibration and standardization preprocessing, multi-dimensional features are extracted through a parallel architecture, fault candidate data are judged through weighted comparison, fault types are recognized after faults are confirmed through continuity verification and abnormality elimination, and the fault types are identified. The scene adaptability is guaranteed by combining regular and instant dual-mode adaptive calibration, and the fault severity can be evaluated and fed back for optimization. The method is high in diagnosis accuracy, good in real-time performance, high in adaptability, high in engineering realizability and suitable for industrial motor operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of motor fault diagnosis technology, specifically to a motor fault diagnosis method based on the inherent characteristic frequency analysis of acceleration signals. It is applicable to diagnostic systems built using conventional electromechanical fault simulation platforms and signal acquisition equipment, enabling real-time identification and location of motor rotor eccentricity, shaft bending, rotor imbalance, and rotor bar faults. It can be widely applied in scenarios such as industrial motor operation and maintenance and intelligent manufacturing equipment monitoring. Background Technology

[0002] As a core power source in industrial production, the operating status of electric motors directly determines production efficiency and safety. According to industrial equipment maintenance data, most motor failures are caused by rotor system anomalies. Failure to diagnose these issues promptly can lead to equipment downtime, production interruptions, and even safety accidents. Therefore, developing efficient and accurate motor fault diagnosis technology has significant engineering value.

[0003] Existing motor fault diagnosis technologies are mainly divided into three categories: First, vibration signal-based diagnostic methods, which identify faults by analyzing the time-domain or frequency-domain characteristics of motor vibration. However, traditional methods often rely on a single feature, and the diagnostic accuracy is greatly affected by operating conditions. Second, current signal-based diagnostic methods, which determine faults through spectrum analysis of motor stator current. However, current signals are easily affected by power grid interference and have low sensitivity to minor faults. Third, auxiliary signal-based diagnostic methods, such as temperature and noise, can only serve as a fault warning reference and cannot locate the fault type.

[0004] Among the diagnostic methods proposed in the published patents, some combine time-domain entropy values ​​with frequency-domain peak features for neural network classification. However, they do not consider the differences in vibration characteristics of different test benches. When the fixed threshold in the laboratory is directly applied to the engineering site, the diagnostic accuracy drops significantly.

[0005] Furthermore, existing technologies generally suffer from a core deficiency of "poor scenario adaptability": different test benches have different motor installation methods, load fluctuation ranges, and environmental noise levels, resulting in significant differences in the vibration characteristics of the same fault under different scenarios. For example, the fundamental frequency amplitude increment of a rotor imbalance fault in a laboratory is within a specific range, but under high-load conditions in an industrial workshop, this increment will change significantly, making it easy for traditional fixed threshold methods to miss the fault. At the same time, existing methods lack sufficient explanation of the fault characteristics mechanism, mostly relying on experimental data statistics and lacking theoretical support, which limits the universality of diagnostic rules. Summary of the Invention

[0006] In view of this, this invention proposes a motor fault diagnosis method based on inherent characteristic frequencies. High-fidelity acceleration signals are acquired using conventional signal acquisition equipment, and the differences in various fault characteristics are clarified by combining motor fault mechanism analysis. A "time-domain-frequency-domain" multi-feature fusion diagnostic model is constructed, and a feature adaptive calibration mechanism is introduced to form a standardized diagnostic process. This solves the problems of weak differentiation of multiple types of rotor faults, poor adaptability to different test bench scenarios, and "fixed thresholds, single features, and ambiguous mechanisms" in existing technologies. It achieves accurate localization of rotor eccentricity, shaft bending, rotor imbalance, and rotor bar faults, adapts to the vibration characteristics of different experimental environments, ensures the repeatability and engineering feasibility of the method, and possesses the technical advantages of short diagnostic delay and high accuracy.

[0007] The specific technical solution is as follows: A motor fault diagnosis method based on inherent characteristic frequencies, the method being implemented using a diagnostic system constructed from an electromechanical fault simulation platform and signal acquisition equipment, includes the following steps: 1) Constructing a normal data feature library: Collect horizontal and vertical acceleration signals of fault-free motors, extract time-domain features, frequency-domain features and feature correlation parameters after preprocessing, establish corresponding feature threshold intervals for the three types of features, and then construct a normal data feature library containing the three types of features and their corresponding threshold intervals. 2) Real-time data acquisition and preprocessing: Acquire bidirectional acceleration signals of the motor to be diagnosed at an appropriate sampling rate, and eliminate noise and errors through filtering, calibration and standardization. 3) Feature extraction and fault candidate determination: The preprocessed data to be diagnosed is segmented at fixed intervals and bidirectional features are extracted and compared with the threshold of the feature library. If any abnormal condition is met, it is determined as fault candidate data. 4) Fault Confirmation: Perform time continuity verification on the fault candidate data. If the verification passes, the fault is confirmed. 5) Fault type localization: Based on the confirmed fault data, identify rotor imbalance, eccentricity, shaft bending, and rotor bar faults through feature comparison; 6) Feature Adaptive Calibration: Update feature library thresholds and fault judgment rules regularly or as needed to ensure scenario adaptability.

[0008] Furthermore, the time-domain features include the acceleration amplitude range and RMS value, the frequency-domain features include the FFT spectrum, the fundamental frequency component amplitude and the amplitude of each harmonic component, and the feature correlation parameters include the correlation between the FFT spectrum in different directions and the normal FFT spectrum. Furthermore, the abnormal conditions include: the correlation between the FFT spectrum in the horizontal or vertical direction and the corresponding FFT spectrum in the normal data feature library is significantly lower than the normal range; the acceleration amplitude in the horizontal or vertical direction is not within the acceleration amplitude range in the corresponding direction in the normal data feature library; and the amplitude of the fundamental frequency component of the FFT in the horizontal or vertical direction is not within the amplitude range of the fundamental frequency component in the corresponding direction in the normal data feature library.

[0009] Furthermore, time continuity verification means that if multiple consecutive sets of data are identified as fault candidate data, and the characteristic deviation value of each set of fault candidate data reaches a set level, and instantaneous abnormal fluctuations are excluded, then the motor is confirmed to have a fault.

[0010] Furthermore, the rules for fault type localization are as follows: a) Rotor imbalance fault: If the amplitude of the acceleration signal in the horizontal and vertical directions is significantly increased compared with the normal data, and the amplitude of the fundamental frequency component in the frequency domain exceeds the upper limit of the amplitude range of the fundamental frequency component in the normal data feature library, then it is determined to be a rotor imbalance fault. b) Rotor eccentricity fault: If the correlation between the vertical FFT spectrum and the normal data FFT spectrum is lower than the conventional threshold, and the amplitude of a certain harmonic component reaches a certain proportion of the fundamental frequency component amplitude, then it is determined to be a rotor eccentricity fault; the certain proportion needs to be set according to the actual situation.

[0011] c) Shaft bending fault: If the correlation between the horizontal FFT spectrum and the normal data FFT spectrum is lower than the conventional threshold, and no harmonic component amplitude is detected to reach a certain proportion of the fundamental frequency component amplitude, then it is determined to be a shaft bending fault; the judgment condition "no harmonic component amplitude is detected to reach a certain proportion of the fundamental frequency component amplitude" is added to distinguish it from rotor eccentricity; the certain proportion needs to be set according to the actual situation.

[0012] d) Rotor bar fault: If the amplitude of the FFT fundamental frequency component is lower than the lower limit of the fundamental frequency component amplitude range in the normal data feature library, and the RMS values ​​of the acceleration signals in both the horizontal and vertical directions are lower than the lower limit of the corresponding RMS value range in the normal data feature library, then it is determined to be a rotor bar fault. Furthermore, the periodic calibration is triggered by time or the number of diagnostics, while the instantaneous calibration is triggered by feature fuzzing or manual intervention, both of which update the global or local feature thresholds, respectively.

[0013] Furthermore, the normal data feature library introduces an environmental compensation factor, and establishes a mapping model by collecting normal motor signals under different operating conditions. During feature comparison, the real-time features are corrected according to real-time environmental parameters.

[0014] Furthermore, it also includes diagnostic result verification and feedback steps, comparing the diagnostic results with the actual fault state, and automatically triggering the calibration process when the accuracy is lower than the threshold.

[0015] Furthermore, the feature extraction adopts a parallel computing architecture, with feature extraction in the horizontal and vertical directions performed simultaneously.

[0016] Compared with existing motor fault diagnosis technologies, the present invention has the following significant advantages: 1. High diagnostic accuracy: Based on a multi-feature fusion scheme of time domain, frequency domain and feature correlation parameters, coupled with clear three types of anomaly judgment conditions and four types of fault-specific location rules, it can accurately distinguish rotor imbalance, eccentricity, shaft bending and rotor bar faults, effectively reducing false judgments and missed judgments, and significantly improving the discrimination ability compared with traditional single feature diagnostic methods. 2. Strong adaptability to different scenarios: By correcting the impact of different working conditions on features through environmental compensation factors, and combining a periodic + real-time dual-mode adaptive calibration mechanism, the feature threshold and judgment rules are dynamically adjusted, which can flexibly adapt to different test benches, working conditions and environmental conditions, and can still maintain stable diagnostic accuracy after application in engineering fields. 3. Excellent real-time performance: It adopts a computing architecture that extracts features in parallel in the horizontal and vertical directions, and processes time-domain and frequency-domain features simultaneously, which greatly reduces the feature extraction time. Combined with clear fault candidate judgment and continuity verification logic, the diagnostic delay is controlled within a reasonable range, meeting the real-time operation and maintenance needs of industrial motors. 4. Continuous optimization of diagnostic reliability: A new diagnostic result verification and feedback mechanism has been added. The accuracy rate is calculated by comparing with the actual fault state. When the accuracy rate is lower than the threshold, the calibration process is automatically triggered to dynamically optimize the feature threshold and fault rules, ensuring the long-term reliability of the diagnostic method. 5. High engineering feasibility: The system is built based on conventional electromechanical fault simulation platforms and signal acquisition equipment. There is no need to customize special hardware. The rules for fault judgment, location and calibration are clear and unambiguous, the diagnostic process is standardized, and it can be solidified into software programs through common programming languages. The development cost is low and it is easy to promote and apply in small and medium-sized enterprises. 6. Strong operation and maintenance guidance: It accurately identifies four types of core rotor faults through clear fault location rules, providing operation and maintenance personnel with clear fault type judgment results, avoiding over-maintenance or under-maintenance due to ambiguous fault types, and reducing operation and maintenance costs. Attached Figure Description

[0017] Figure 1 A flowchart illustrating one implementation of the method described in this invention; Figure 2 Diagnostic system architecture diagram of the present invention; Figure 3 The flowchart of the method described in this invention. Detailed Implementation

[0018] The core technical solution of this invention is a three-level architecture of "feature library construction - real-time diagnosis - adaptive calibration". Based on the acceleration signal with an adaptive sampling rate, fault identification and location are achieved through multi-feature fusion. Specifically, it includes experimental platform construction, diagnostic system architecture and core diagnostic process.

[0019] This diagnostic method is based on a comprehensive experimental platform for simulating electromechanical faults, which forms a complete diagnostic system. The system uses a hardware platform as support, functional units as the core, and standardized processes as the link to achieve accurate diagnosis of motor faults, as detailed below: I. Experimental Platform Setup A conventional electromechanical fault simulation experimental platform was adopted, equipped with a compatible asynchronous motor, and a signal acquisition unit and an acceleration sensor were configured. The sensor was precisely installed at the horizontal and vertical radial positions of the front bearing housing of the motor to ensure that the acquired signal could accurately reflect the radial vibration state of the rotor. The sampling accuracy, input range, triggering method, and data transmission method of the signal acquisition unit were all adapted to the diagnostic requirements. The sampling parameters were set to adapted values, and the acquired bidirectional acceleration signals were transmitted to an industrial computer in real time. The data processing software deployed in the industrial computer integrated core functional modules such as filtering, FFT analysis, feature calculation, and data storage, providing basic support for data processing, diagnostic analysis, and backtracking queries.

[0020] II. Diagnostic System Architecture The diagnostic system is based on experimental platform hardware and is divided into four main functional units. Each unit interacts with data through standardized data interfaces to ensure compatibility and scalability. (1) Data acquisition unit: It consists of a conventional electromechanical fault comprehensive simulation experimental platform and a signal acquisition unit. The simulation platform is used to simulate the normal and four fault motor operating states and supports the installation of horizontal / vertical direction sensors; the signal acquisition unit is used to synchronously acquire horizontal and vertical acceleration signals, and the sampling parameters are set to the adaptation value to ensure that no signal is lost.

[0021] (2) Data processing unit: including signal preprocessing module and feature extraction module. The preprocessing module uses Butterworth filtering of appropriate order (passband frequency covers the fault feature range), signal calibration (amplitude error is controlled at a low level) and Z-score normalization; the feature extraction module calculates time domain features and frequency domain features in parallel, and the calculation time is controlled within a reasonable range.

[0022] (3) Diagnostic decision unit: including fault judgment module and fault location module. The fault judgment module confirms the fault by comparing features over a certain period of time; the fault location module distinguishes the fault type based on the feature priority mechanism and introduces feature matching degree calculation to solve the feature cross problem.

[0023] (4) Calibration unit: It realizes two modes of periodic calibration and instant calibration. It collects current scene samples through the simulation platform and updates the threshold of the feature library.

[0024] III. Core Diagnostic Process Step S1: Construct a scenario-based normal data feature library A fault-free motor was run on the target test bench (simulation platform) to collect samples under multiple operating conditions, covering different load levels and speed levels. The motor was run for a sufficient amount of time under each operating condition, and the data collection time for each set was set to a fixed value. The acceleration signals of the motor in the horizontal and vertical directions were collected synchronously through a signal acquisition device, and the sampling rate was set to an adaptive value.

[0025] After preprocessing the acquired normal signals, the data is segmented according to a fixed period, and a data validity judgment is performed simultaneously: only sample data with a high signal-to-noise ratio, small amplitude fluctuation range, and no obvious sudden noise peaks in the FFT spectrum are retained. A high success rate for acquiring valid data is required; samples that do not meet the criteria must be re-acquired. For the filtered valid data, statistical values ​​of time-domain features (acceleration amplitude, RMS value), frequency-domain features (fundamental frequency component amplitude), and FFT spectrum correlation are calculated, and the normal threshold range for each feature is determined using the following method: (1) Time-domain features (acceleration amplitude, RMS value): Calculate the mean μ and standard deviation σ of the corresponding features of all normal samples, and set a reasonable threshold range based on μ and σ; (2) Frequency domain characteristics (fundamental frequency component amplitude): After removing outlier samples using outlier removal criteria, the characteristic extreme values ​​of the remaining samples are calculated, and a reasonable threshold range is set based on these extreme values. (3) FFT spectral correlation: By calculating the correlation distribution between different groups within the normal sample data, the lower limit of the high confidence interval is taken as the normal correlation threshold.

[0026] Simultaneously, normal motor acceleration signals under different ambient temperatures and humidity levels are collected to establish a mapping model between environmental parameters and characteristic deviations. An environmental compensation factor (within a small positive and negative range) is generated and included in the feature library. During subsequent feature comparison steps, the corresponding compensation factor can be called to correct the real-time features based on the real-time environmental parameters. The correction formula is: xcorrected = xreal-time × (1 + compensation factor). The feature library's storage format must support data backtracking analysis to ensure that data can be queried and verified during subsequent diagnostic and calibration processes. This example only uses different ambient temperatures and humidity levels, but is not limited to the aforementioned temperature and humidity conditions.

[0027] Step S2: Real-time data acquisition and preprocessing The accelerometer is fixed in the horizontal and vertical positions of the bearing housing at the front end of the motor to ensure that the acquired signal can accurately reflect the rotor vibration characteristics. The sampling parameters of the signal acquisition device are set according to the diagnostic requirements. Its sampling accuracy, input range, triggering method and data transmission method must all match the diagnostic requirements to ensure the integrity and real-time performance of the acquired data and avoid feature extraction errors caused by data loss.

[0028] The signal acquisition unit continuously acquires bidirectional acceleration signals of the motor to be diagnosed at an appropriate sampling rate, and outputs the acquired data to the processing unit at a fixed period. The following preprocessing procedure is performed on the output real-time signal to eliminate the influence of environmental noise and equipment error on the diagnostic results.

[0029] Sub-step S21: Signal filtering. The acquired signal is filtered using a Butterworth filter of appropriate order. The passband frequency of the filter is set to cover the range of motor fault characteristics, and low-frequency interference and high-frequency noise are accurately filtered out. Sub-step S22: Signal calibration. The signal acquisition equipment is calibrated using a standard vibration source to control the amplitude error of the acquired signal to a low level and ensure signal accuracy. Sub-step S23: Data standardization. The Z-score standardization method is used to process the filtered signal to eliminate signal amplitude drift caused by different acquisition periods and changes in ambient temperature. The standardization formula is: x'=(x-μ x ) / σ x Where x is the original signal value, μ x σ is the mean value of the signal within this acquisition period. x This represents the standard deviation of the signal within the acquisition period.

[0030] Step S3: Parallel extraction and real-time judgment of multiple features A parallel computing architecture is used to extract features in both horizontal and vertical directions simultaneously. The specific features and extraction methods are as follows: (1) Time domain characteristics: ① Amplitude range: calculate the maximum and minimum values ​​of each set of data; ② RMS value: calculated according to the standard formula, reflecting the effective value of vibration energy.

[0031] (2) Frequency domain characteristics: ① FFT spectrum: After adding a Hanning window to each set of data, perform FFT transformation to obtain the frequency domain spectrum of the corresponding range; ② Fundamental frequency identification: Obtain the motor speed, calculate the theoretical fundamental frequency, and search for the actual fundamental frequency and corresponding amplitude within a reasonable range near the theoretical fundamental frequency; ③ Harmonic amplitude ratio: Calculate the ratio of each harmonic amplitude to the fundamental frequency amplitude, and focus on specific harmonics related to rotor eccentricity.

[0032] (3) Feature association parameter: FFT correlation, which uses Pearson correlation coefficient to calculate the similarity between the real-time FFT spectrum and the normal FFT spectrum of the feature library.

[0033] Real-time judgment rules: After environmental compensation, the extracted features are compared with the threshold of the feature library. If any of the following conditions are met, the data is judged as a candidate for fault: ① The correlation is significantly lower than the threshold; ② The amplitude / RMS exceeds the reasonable range; ③ The fundamental frequency amplitude exceeds the set range. At the same time, a weighted feature deviation value is introduced for calculation (correlation has the highest weight, followed by fundamental frequency, and amplitude and RMS have relatively lower weights). When the weighted deviation reaches a certain threshold, the judgment of fault candidates is strengthened.

[0034] Step S4: Continuous verification and fault confirmation Considering the continuous nature of motor faults, fault candidate data is verified continuously for a certain period of time. An abnormal fluctuation exclusion mechanism is introduced: if a set of fault candidate data is present, but adjacent sets are normal and have small characteristic deviations, it is judged as transient interference and not included in the count; only when multiple consecutive sets are valid fault candidate data is the motor fault confirmed and an alarm triggered.

[0035] Step S5: Fault location, fault simulation, and diagnostic verification Fault localization is the core step in the diagnostic process of this invention. Pre-defined diagnostic rules are derived based on the essential mechanisms of motor faults, and combined with specific handling strategies for feature intersections and complex faults, accurate identification of fault types is achieved. Subsequent fault simulation experiments verified the effectiveness and reliability of the rules, as detailed below: (I) Theoretical support for fault characteristic mechanism The fault diagnosis rules of this invention are derived from an in-depth analysis of the nature of motor faults, and the inherent correlation between various fault characteristics and mechanisms provides a solid theoretical basis for the location rules: Rotor imbalance fault: Low horizontal correlation + no significant harmonics. When the rotor mass distribution is uneven, the centrifugal force F=mrω² (m is the eccentric mass, r is the eccentricity, and ω is the angular velocity) generated by rotation increases with the square of the rotational speed, resulting in synchronous enhancement of vibration in both the horizontal and vertical directions, and a significant increase in the time-domain amplitude. In the frequency domain, the vibration energy is concentrated at the fundamental frequency (consistent with the rotational speed), and the amplitude of the fundamental frequency component significantly exceeds the normal range. Relevant technical literature and experimental data verify that the bidirectional amplitude and fundamental frequency amplitude of this type of fault are significantly enhanced compared to the normal state, providing a direct basis for diagnostic rules.

[0036] Rotor eccentricity fault: When the rotor and stator centers do not coincide, the symmetry of the stator magnetic field is disrupted, generating unilateral magnetic pull (radial force) and exciting slot harmonics (frequency related to the number of pole pairs and speed of the motor). Most motors are horizontally mounted, and the vertical bearing stiffness is lower than that in the horizontal direction, making them more sensitive to eccentric vibrations. This results in a significant difference between the vertical FFT spectrum and the normal spectrum (reduced correlation). Simultaneously, the amplitude of specific harmonic components (selected based on the number of pole pairs and the distortion law of the air gap magnetic field) increases with the degree of eccentricity. Related studies show that the ratio of harmonic amplitude to fundamental frequency amplitude is in a high range during eccentricity faults, supporting the rationality of the characteristic rules.

[0037] Shaft bending fault: Shaft bending causes the rotor's rotation center trajectory to become elliptical, resulting in centrifugal force frequencies concentrated at the fundamental frequency (similar to unbalance faults). However, the eccentricity caused by bending is fixed and will not excite slot harmonics. Horizontal bearing supports of horizontally mounted motors are more susceptible to bending deformation, leading to a significant decrease in the correlation between the horizontal FFT spectrum and the normal spectrum, and the absence of excessive harmonic component amplitudes. Relevant technical standards (such as GB / T 10068-2018) clearly define the spectral characteristics of this type of fault as "prominent fundamental frequency, no additional harmonics," which perfectly matches the rules of this invention.

[0038] Rotor bar faults: Broken or loose rotor bars lead to increased rotor circuit impedance, decreased rotor current, reduced air gap magnetic field amplitude, weakened electromagnetic excitation on the stator, and a decrease in overall vibration energy. In the time domain, this manifests as a lower RMS acceleration value (effective value of vibration energy), and in the frequency domain, a synchronous decrease in the amplitude of the fundamental frequency component. Relevant experimental data are consistent with the simulation results of this invention, verifying the necessary correlation between this characteristic and the fault.

[0039] (II) Fault Location After confirming a motor fault in step S4, based on the preset rules derived from the above mechanism and combined with feature detail analysis, the fault type is accurately located, and specific handling strategies are formulated for special scenarios: 1. Single Fault Location Rules Rotor imbalance fault: The diagnostic rule is that the acceleration amplitude in both the horizontal and vertical directions significantly exceeds the upper limit of the corresponding interval in the normal data feature library, and the amplitude of the fundamental frequency component in the frequency domain significantly exceeds the upper limit of the normal interval.

[0040] Rotor eccentricity fault: The diagnostic rule is that the correlation between the vertical FFT spectrum and the normal data FFT spectrum is lower than the conventional threshold, and the amplitude of a certain harmonic component (preferably a specific harmonic, selected according to the number of motor pole pairs and the distortion law of the air gap magnetic field) reaches a certain proportion of the fundamental frequency component amplitude; Special scenario handling: If no significant increase of the preferred harmonic component is detected, other related harmonic components are checked in turn. If the amplitude ratio requirement is met and the vertical correlation is lower than the threshold, it is still determined to be this fault.

[0041] Shaft bending fault: The diagnostic rule is that the correlation between the horizontal FFT spectrum and the normal data FFT spectrum is lower than the conventional threshold, and the amplitudes of all harmonic components do not reach a certain proportion of the fundamental frequency component amplitude, while the vertical correlation is within the normal range.

[0042] Rotor bar fault: The diagnostic rule is that the amplitude of the FFT fundamental frequency component is significantly lower than the lower limit of the corresponding interval in the normal data feature library, and the RMS values ​​of the acceleration signals in both the horizontal and vertical directions are significantly lower than the lower limit of the normal interval. 2. Special Scenario Handling Strategies Feature cross-processing: When fault data simultaneously satisfies some features of multiple single faults, a feature priority determination mechanism is adopted (priority from high to low: harmonic features → horizontal FFT correlation → fundamental frequency amplitude and RMS correlation features → bidirectional amplitude features). When high-priority features are satisfied, the corresponding fault is directly determined; if feature cross-processing still exists, the feature matching degree of each fault type is calculated (calculation formula: matching degree = number of satisfied features / total number of features of the fault × 100%), and the one with the highest matching degree is taken as the final diagnosis result.

[0043] Complex Fault Handling: For scenarios where two or more faults coexist in the engineering field, the Independent Component Analysis (ICA) algorithm is used to decompose the FFT spectrum of the complex fault into the feature spectrum components of each individual fault. Then, the components are compared with the feature rule set of the corresponding individual fault to clarify the composition type of the complex fault.

[0044] For special scenarios, further analysis is needed based on the actual situation.

[0045] (III) Fault Simulation and Diagnostic Verification To verify the accuracy, reliability, and adaptability of the fault location method, a special experiment was conducted on a conventional electromechanical fault simulation platform. Four single faults and a combined fault of "rotor imbalance + eccentricity" were simulated. Each fault was run for a sufficient time under specific load and speed conditions, and horizontal / vertical acceleration signals were collected synchronously. The complete diagnostic process of this invention was followed for testing, and the diagnostic results, feature matching degree, and accuracy were recorded, as detailed below: Rotor imbalance fault diagnosis Fault characteristics: The amplitudes of both horizontal and vertical accelerations significantly exceed the normal upper limit, the amplitude of the fundamental frequency component significantly exceeds the upper limit, the correlation of the FFT spectrum is lower than the normal threshold, and multiple consecutive sets of data meet the fault candidate conditions and trigger an alarm. Diagnostic results: The characteristics perfectly match the rotor imbalance fault rules, the characteristic matching degree reaches a high level, and the severity is accurately graded according to the amplitude ratio; Verification results: The diagnostic accuracy reached a high level, with no misjudgments or omissions, verifying the consistency between the fault location rules and the mechanism analysis.

[0046] Rotor eccentricity fault diagnosis Fault characteristics: Vertical FFT spectrum correlation is lower than the normal threshold, the ratio of the amplitude of a specific harmonic component to the fundamental frequency amplitude reaches the set ratio, horizontal correlation is within the normal range, and multiple consecutive sets of data meet the fault candidate conditions. Diagnostic results: The characteristics and rotor eccentricity fault rules are precisely matched, the characteristic matching degree reaches a high level, and the severity is reasonably graded according to the harmonic ratio; Verification results: The diagnostic accuracy reached a high level. Potential misjudgments caused by occasional transient noise could be corrected through the feature-adaptive calibration process. No confusion with other faults occurred, verifying the effectiveness of the harmonic component identification rules.

[0047] Shaft bending fault diagnosis Fault characteristics: The horizontal FFT spectrum correlation is lower than the normal threshold, the amplitude ratio of all harmonic components is lower than the set level, the vertical correlation is within the normal range, the fundamental frequency amplitude is slightly higher than the normal upper limit, and multiple consecutive sets of data meet the fault candidate conditions. Diagnostic results: The characteristics highly match the shaft bending fault rules, the characteristic matching degree reaches a high level, and the severity is scientifically graded according to the correlation difference. Verification results: The diagnostic accuracy reached a high level, successfully distinguishing it from rotor eccentricity fault (the core characteristic of no excessive harmonic amplitude), verifying the rule's distinguishing ability and the rationality of its mechanism support.

[0048] Rotor bar fault diagnosis Fault characteristics: The amplitude of the fundamental frequency component is significantly lower than the normal lower limit, the RMS values ​​of both horizontal and vertical accelerations are significantly lower than the normal lower limit, the correlation of the FFT spectrum is lower than the threshold, and multiple consecutive sets of data meet the fault candidate conditions; Diagnostic results: The characteristics match the rotor bar fault rules, the characteristic matching degree reaches a high level, and the severity is accurately graded according to the ratio of fundamental frequency amplitude; Verification results: The diagnostic accuracy reached a high level. Temporary anomalies caused by load fluctuations could be verified and corrected through subsequent continuous data. The stability was good, and the correlation between the vibration energy reduction characteristics and the fault mechanism was verified.

[0049] Composite Fault Diagnosis Verification Fault simulation: Construct a composite fault scenario of "rotor imbalance + eccentricity", in which the fault characteristics overlap; Processing procedure: The ICA algorithm was used to separate the features of the FFT spectrum of the composite fault, and the feature spectrum components of the two single faults, rotor imbalance and rotor eccentricity, were successfully decomposed. Verification results: The separated feature components were accurately matched with the corresponding single fault rules, and the final diagnosis results were consistent with the fault settings of the simulation platform, verifying the effective localization capability of the present invention for complex faults.

[0050] The above examples are just illustrations of complex faults; specific judgments should be made based on the actual situation.

[0051] In summary, the fault location method of this invention is based on clear fault mechanism construction rules and solves the problem of feature intersection and compound faults through special scenario processing strategies. Experimental verification shows that it can achieve accurate diagnosis in both single and compound fault scenarios. The diagnostic accuracy rate under standard operating conditions reaches a high level, and the false positive rate and false negative rate are controlled at a low level. Moreover, it can correct the anomalies caused by field interference through calibration, continuous verification and other mechanisms, and is fully adapted to the actual engineering application needs.

[0052] Step S6: Fault Severity Assessment After fault location, the severity of the fault is quantified based on characteristic deviations, providing a basis for operation and maintenance: (1) Imbalance: The ratio of the mean amplitude in both directions to the normal mean, classified as mild, moderate and severe according to the size of the ratio; (2) Eccentricity: Specific harmonic ratios are classified into mild, moderate and severe according to their magnitude; (3) Curvature: The absolute value of the difference between the horizontal correlation and the conventional threshold, which is divided into mild, moderate and severe according to the size of the difference; (4) Rotor bar: The ratio of the fundamental frequency amplitude to the normal lower limit value, classified as mild, moderate and severe according to the size of the ratio.

[0053] Step S7: Feature Adaptive Calibration To address the differences between different test benches, two calibration modes are employed: (1) Periodic calibration: After running for a certain period of time or accumulating a certain number of diagnoses, the normal and four standard fault motors are run sequentially through the simulation platform. Sufficient data is collected for each state, and the feature library threshold and fault rule parameters are recalculated. (2) Real-time calibration: When the matching degree is low for multiple consecutive diagnoses or when the user triggers it, only samples related to the current working condition are collected, and the relevant thresholds are locally adjusted.

[0054] The method of this invention was applied to different models of motors mounted on another test bench. Before calibration, the diagnostic accuracy was low. After initiating a periodic calibration process, normal and fault data were collected using the test bench, and the feature library thresholds were updated. After calibration, the diagnostic accuracy significantly improved, verifying the scene adaptation capability.

[0055] Step S8: Verification and Feedback of Diagnostic Results In laboratory settings, diagnostic results are compared with the fault settings of the simulation platform; in engineering settings, they are compared with the results of disassembly and inspection to calculate the diagnostic accuracy. When the accuracy falls below a higher threshold, a calibration process is automatically triggered, recording fault types and operating conditions with significant deviations, and optimizing the feature rule set.

[0056] Experimental verification shows that this method achieves a high level of accuracy in diagnosing four types of single and compound faults, with diagnostic delays controlled within a short time. After scenario calibration, the accuracy for on-site engineering adaptation remains at a high level. This invention requires no custom hardware, integrates fault mechanism and multi-feature analysis, and can be widely applied to industrial motor maintenance and intelligent manufacturing equipment monitoring scenarios.

[0057] The scope of protection of this invention is not limited to the specific embodiments described above. For those skilled in the art, adjustments can be made to the details of each step without departing from the principle of this invention, and such adjustments all fall within the scope of protection of this invention.

Claims

1. A motor fault diagnosis method based on inherent characteristic frequency, characterized in that, The method is implemented based on a diagnostic system constructed using an electromechanical fault simulation platform and signal acquisition equipment, and includes the following steps: 1) Constructing a normal data feature library: Collect horizontal and vertical acceleration signals of fault-free motors, extract time-domain features, frequency-domain features and feature correlation parameters after preprocessing, establish corresponding feature threshold intervals for the three types of features, and then construct a normal data feature library containing the three types of features and their corresponding threshold intervals. 2) Real-time data acquisition and preprocessing: Acquire bidirectional acceleration signals of the motor to be diagnosed at an appropriate sampling rate, and eliminate noise and errors through filtering, calibration and standardization. 3) Feature extraction and fault candidate determination: The preprocessed data to be diagnosed is segmented at fixed intervals and bidirectional features are extracted and compared with the threshold of the feature library. If any abnormal condition is met, it is determined as fault candidate data. 4) Fault Confirmation: Perform time continuity verification and anomaly elimination on the candidate fault data. If the verification is successful, the fault is confirmed. 5) Fault type localization: Based on the confirmed fault data, identify rotor imbalance, eccentricity, shaft bending, and rotor bar faults through feature comparison; 6) Feature Adaptive Calibration: Update feature library thresholds and fault judgment rules regularly or as needed to ensure scenario adaptability.

2. The method according to claim 1, characterized in that, The time-domain features include the acceleration amplitude range and RMS value, the frequency-domain features include the FFT spectrum, the fundamental frequency component amplitude and the amplitude of each harmonic component, and the feature correlation parameters include the correlation between the FFT spectrum in different directions and the normal FFT spectrum.

3. The method according to claim 1, characterized in that, The abnormal conditions include: the correlation between the FFT spectrum in the horizontal or vertical direction and the corresponding FFT spectrum in the normal data feature library is significantly lower than the normal range; the acceleration amplitude in the horizontal or vertical direction is not within the acceleration amplitude range in the corresponding direction in the normal data feature library; and the amplitude of the fundamental frequency component of the FFT in the horizontal or vertical direction is not within the amplitude range of the fundamental frequency component in the corresponding direction in the normal data feature library.

4. The method according to claim 1, characterized in that, Confirming a fault specifically means that multiple consecutive sets of data are identified as fault candidate data, and the characteristic deviation value of each set of fault candidate data reaches a set level, then the motor is confirmed to have a fault.

5. The method according to claim 1, characterized in that, The rules for fault type location are as follows: a) Rotor imbalance fault: If the amplitude of the acceleration signal in the horizontal and vertical directions is significantly increased compared with the normal data, and the amplitude of the fundamental frequency component in the frequency domain exceeds the upper limit of the amplitude range of the fundamental frequency component in the normal data feature library, then it is determined to be a rotor imbalance fault. b) Rotor eccentricity fault: If the correlation between the vertical FFT spectrum and the normal data FFT spectrum is lower than the conventional threshold, and the amplitude of a certain harmonic component reaches a certain proportion of the fundamental frequency component amplitude, then it is determined to be a rotor eccentricity fault. c) Shaft bending fault: If the correlation between the horizontal FFT spectrum and the normal data FFT spectrum is lower than the conventional threshold, and no harmonic component amplitude is detected to reach a certain proportion of the fundamental frequency component amplitude, then it is determined to be a shaft bending fault. d) Rotor bar fault: If the amplitude of the FFT fundamental frequency component is lower than the lower limit of the fundamental frequency component amplitude range in the normal data feature library, and the RMS values ​​of the acceleration signals in both the horizontal and vertical directions are lower than the lower limit of the corresponding RMS value range in the normal data feature library, then it is determined to be a rotor bar fault.

6. The method according to claim 1, characterized in that, The periodic calibration is triggered by time or the number of diagnostics, while the instantaneous calibration is triggered by feature fuzzing or manual intervention, which respectively update the global or local feature thresholds.

7. The method according to claim 1, characterized in that, The normal data feature library introduces an environmental compensation factor, and establishes a mapping model by collecting normal motor signals under different operating conditions. During feature comparison, the real-time features are corrected according to the real-time environmental parameters.

8. The method according to any one of claims 1-7, characterized in that, It also includes diagnostic result verification and feedback steps, comparing the diagnostic results with the actual fault status, and automatically triggering the calibration process when the accuracy is lower than the threshold.

9. The method according to claim 8, characterized in that, The feature extraction adopts a parallel computing architecture, with feature extraction in the horizontal and vertical directions performed simultaneously.

Images