Grinding machine internal part temperature anomaly detection method based on vibration signal analysis

Abstract

The invention discloses a grinding machine internal part temperature anomaly detection method based on vibration signal analysis, which comprises the following steps that grinding data are collected through sensor deployment, and the sensors comprise a temperature sensor, a vibration sensor, an infrared thermal imaging sensor, a magnetic resistance current sensor and an inductance type particle sensor; carrying out preprocessing and feature extraction on the collected data; model construction and training are carried out based on data of preprocessing and feature extraction; according to the method, the abnormal condition of the temperature of the part is predicted by detecting the vibration signal of the internal part, the content of metal particles in lubricating oil is detected through the oil analysis sensor, and the abrasion degree of the bearing is judged in combination with the vibration signal. And motor current harmonic characteristics are monitored, and overload or rotor imbalance problems are identified.

Description

Technical Field

[0001] The present invention relates to the technical field of detection, and specifically to a method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis. Background Technique

[0002] At present, the detection of abnormal internal parts of a grinding machine mainly relies on single temperature or vibration threshold monitoring. For example: Temperature monitoring: The temperature of key parts is collected in real time through temperature sensors, and an alarm is triggered when the temperature exceeds the preset threshold. Vibration monitoring: The vibration amplitude is detected by an acceleration sensor, and early warning is carried out when the vibration is abnormal.

[0003] The defects existing in the prior art are: Single parameter dependence: Traditional methods only monitor the temperature threshold or vibration amplitude, and cannot capture the correlation characteristics between multiple parameters. For example, when the temperature exceeds the threshold and an alarm is given, the fault may have entered an irreversible stage, causing damage to equipment parts and thus economic losses. Isolated analysis: Early warning is carried out by simply measuring the temperature, without considering the dynamic correlation between temperature and vibration, and without establishing a mapping relationship between vibration spectrum changes and temperature anomalies, resulting in difficulty in identifying compound faults such as heat generation due to friction. High false alarm rate: A single temperature signal is easily interfered by working condition fluctuations and cannot distinguish normal fluctuations from real faults. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides a method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis. The present invention predicts abnormal part temperature by detecting the vibration signal of internal parts, detects the metal particle content in the lubricating oil through an oil analysis sensor, and combines the vibration signal to judge the bearing wear degree. Monitor the harmonic characteristics of the motor current to identify overload or rotor imbalance problems. In this way, early warning of early faults of internal parts of the equipment is realized, the risk of equipment shutdown is reduced, and economic losses are reduced.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis, comprising the following steps:

[0007] Step S1: Collect grinding data through sensor deployment, where the sensors are temperature sensors, vibration sensors, infrared thermal imaging sensors, magnetoresistive current sensors, and inductive particle sensors;

[0008] Specifically: Temperature sensor: Used to directly measure the temperature of key parts inside the grinding machine, such as the temperature changes at positions such as bearings and grinding chambers, and provide direct data for abnormal temperature detection.

[0009] Triaxial vibration sensor: Installed at positions such as the outer shell and bearing housing of the grinding machine to collect vibration signals during the grinding process. Different changes in the internal part states will cause vibrations with different characteristics. For example, part looseness, wear, etc. will cause changes in parameters such as vibration amplitude and frequency, and these changes may be related to abnormal temperature.

[0010] Infrared thermal imaging sensor: Can perform non-contact temperature measurement on the whole grinding machine or specific areas, obtain the surface temperature distribution image, which helps to discover local temperature abnormal areas. Especially for some parts where it is not easy to install temperature sensors or areas with rapid temperature changes, it can provide more comprehensive temperature information.

[0011] Magnetoresistive current sensor: By monitoring the current change of the grinding machine motor, it indirectly reflects the load condition of the grinding machine. When there is an abnormal temperature in the internal parts, it may cause a change in the motor load, and then cause fluctuations in the current. The magnetoresistive current sensor can capture these subtle changes.

[0012] Inductive particle sensor: Mainly used to detect information such as particle concentration and size generated during the grinding process. If the wear of internal parts intensifies, it will cause an increase in the number of generated particles or a change in particle size, which may be related to the abnormal increase in the temperature of the parts, because high temperature may accelerate the wear of the parts.

[0013] Step S2: Preprocess and extract features from the collected data;

[0014] Specifically: Denoising: Use filtering algorithms such as Gaussian filtering and median filtering to remove noise in the collected data, improve the quality and stability of the data, and reduce the interference of noise on subsequent analysis.

[0015] Normalization: Normalize the data collected by different sensors with different dimensions and ranges so that they are within the same numerical range, which is convenient for subsequent feature extraction and model training.

[0016] Feature extraction:

[0017] Time-domain features: Calculate statistical features such as the mean, variance, peak value, and kurtosis of the vibration signal. These features can reflect the intensity and stability of the vibration; for the temperature signal, extract features such as the average value, maximum value, and minimum value of the temperature to describe the change range and trend of the temperature.

[0018] Frequency-domain features: Perform frequency-domain analysis such as Fourier transform on the vibration signal to obtain the frequency components of the signal, and extract features such as the main frequency and amplitude spectrum. Different fault types often generate characteristic frequencies in specific frequency bands, which helps to identify potential causes of abnormal temperature.

[0019] Time-frequency domain features: By using time-frequency analysis methods such as wavelet transform, the signal is analyzed in both the time and frequency domains simultaneously to extract time-frequency distribution features, which can capture the changes of the signal at different times and frequencies more precisely and have better analysis effects on complex and non-stationary grinding process signals.

[0020] Step S3: Based on the data of preprocessing and feature extraction, model construction and training are carried out.

[0021] Specifically, select a suitable model: According to the characteristics and requirements of the abnormal temperature detection of internal parts of the grinding machine, select a suitable machine learning or deep learning model. Common models include support vector machine (SVM), artificial neural network (ANN), random forest (RF), etc. For example, the artificial neural network has a powerful non-linear mapping ability and can handle complex input-output relationships, which is suitable for multi-parameter and non-linear systems such as grinding machines. Divide the dataset: Divide the data after preprocessing and feature extraction into a training set, a validation set, and a test set. The training set is used to train the parameters of the model, the validation set is used to adjust the hyperparameters of the model and prevent overfitting, and the test set is used to evaluate the performance and generalization ability of the model. Model training and optimization: Use the training set to train the model. By adjusting the parameters of the model, such as weights and biases in the neural network, the model can minimize the loss function, such as mean square error (MSE), cross-entropy, etc. During the training process, optimization algorithms such as stochastic gradient descent (SGD), Adagrad, Adadelta, etc. can be used to update the parameters of the model to accelerate the convergence speed of the model. At the same time, adjust the hyperparameters of the model according to the results of the validation set, such as learning rate, number of iterations, number of hidden layer nodes, etc., to obtain the optimal model performance.

[0022] Step S4: Implement abnormal detection and warning according to the constructed model.

[0023] Specifically, set a threshold: According to the output result of the model, combined with the actual operating conditions and historical data of the grinding machine, set a reasonable abnormal threshold. For example, when the abnormal temperature probability output by the model exceeds a certain threshold, it is determined that a temperature abnormal situation occurs. Abnormal detection: After preprocessing and feature extraction of the real-time collected data, input it into the trained model for prediction. The model will output the temperature status information of the current internal parts of the grinding machine, such as whether there is an abnormality and the degree of abnormality. Warning mechanism: When an abnormal situation is detected, a warning signal is sent in a timely manner. The warning methods can include sound alarm, light flashing, SMS notification, email reminder, etc., so that the operator can take corresponding measures in a timely manner, such as stopping for inspection, adjusting process parameters, etc., to avoid the further expansion of the fault and ensure the safe and stable operation of the grinding machine. At the same time, record and analyze the abnormal data to provide a reference basis for subsequent fault diagnosis and equipment maintenance.

[0024] As a further solution of the present invention, in the step S1, the specific data acquisition process is as follows: temperature sensors are installed on the bearings and gearboxes of the grinding machine to collect the current temperature signals of the parts, and the sampling frequency is ≥1 Hz; vibration sensors are installed to capture the high-frequency vibration signals of the parts, and the sampling frequency is ≥5 kHz; infrared thermal imaging sensors are installed to monitor the heat generation of the internal parts and realize the monitoring of the spatial temperature distribution; magnetoresistive current sensors are used to monitor the harmonic characteristics of the motor current to identify overload or rotor imbalance; inductive particle sensors are used to measure the metal particle content in the lubricating oil, and the bearing wear degree is judged in combination with the vibration signals.

[0025] Further preferably, for the temperature sensors: temperature sensors are installed on the bearings and gearboxes of the grinding machine to collect the current temperature signals of the parts at a sampling frequency of ≥1 Hz. Such a sampling frequency can capture the temperature changes in a relatively timely manner and is helpful for detecting abnormal temperature fluctuations. For example, when a bearing or gearbox fails, the temperature may gradually increase. Through continuous temperature data acquisition, the rising trend of the temperature can be observed, thus giving an early warning of potential problems.

[0026] For the vibration sensors: vibration sensors are installed to capture the high-frequency vibration signals of the parts, and the sampling frequency is ≥5 kHz. The higher sampling frequency is to accurately obtain the detailed information of the part vibration. Because during the operation of the grinding machine, the vibration of the parts is relatively complex and contains various frequency components. The high-frequency vibration signals often hide the characteristic information of part failures. For example, the wear of the rolling elements of the bearing and the tooth surface damage of the gear will cause vibration changes at specific frequencies. By analyzing these high-frequency vibration signals, small part failures can be detected in time to avoid the further expansion of the failures.

[0027] For the infrared thermal imaging sensors: infrared thermal imaging sensors are used to monitor the heat generation of the internal parts and realize the monitoring of the spatial temperature distribution. It can obtain the temperature field distribution image of the internal parts of the grinding machine in a non-contact manner and intuitively show which parts have overheating phenomena. For example, when abnormal friction occurs at the meshing part of a gear, the temperature of that part will rise, and the infrared thermal imaging sensor can clearly capture this hot spot, helping the maintenance personnel quickly locate the fault position. Moreover, by comparing the thermal imaging images at different times, the temperature change trend can also be analyzed to provide more comprehensive information for fault diagnosis.

[0028] Magnetoresistive current sensor: A magnetoresistive current sensor is used to monitor the harmonic characteristics of the motor current to identify problems such as overload or rotor imbalance. When the motor is operating normally, the waveform and frequency of the current are relatively stable. When an overload occurs, the current increases and the harmonic components also change; while rotor imbalance causes periodic fluctuations in the motor current. The magnetoresistive current sensor can accurately measure the harmonic characteristics of the current. By analyzing these characteristics, it can be determined whether the motor is operating normally, potential electrical faults can be detected in a timely manner, and the motor can be prevented from being damaged due to overload or imbalance.

[0029] Inductive particle sensor: An inductive particle sensor is used to measure the metal particle content in the lubricating oil. Combining with vibration signals can determine the wear degree of the bearing. During the operation of the grinding machine, the wear of the bearing causes metal particles to fall off and mix into the lubricating oil. The inductive particle sensor can detect the number and size of metal particles in the lubricating oil. When the particle content increases, it indicates that the bearing may have relatively serious wear. At the same time, combining with vibration signals can further confirm the wear condition of the bearing. For example, when characteristic frequencies related to bearing faults appear in the vibration signal and the metal particle content in the lubricating oil is also high, it can be more accurately determined that the bearing has wear problems and timely maintenance or replacement is required.

[0030] As a further solution of the present invention, in step S2, the data preprocessing includes data cleaning, frequency band division and energy calculation, temperature feature extraction, motor current harmonic feature extraction, and lubricating oil metal particle content feature extraction.

[0031] Further, data cleaning:

[0032] Removing noise: The collected data may contain various noises, such as the electronic noise of the sensor itself, electromagnetic interference noise in the environment, etc. Digital filters, such as low-pass filters, high-pass filters, band-pass filters, or wavelet denoising, can be used to remove noise according to the frequency characteristics of the noise and retain the useful signal components. For example, if it is known that the noise in the vibration signal is mainly high-frequency noise, a low-pass filter can be used to filter out the high-frequency part.

[0033] Filling missing values: During the data acquisition process, some data may be missing due to sensor failures, data transmission problems, etc. For missing values, methods such as mean filling, median filling, time-series-based interpolation methods (such as linear interpolation, spline interpolation), or machine learning-based filling methods (such as K-nearest neighbor filling) can be used for filling. For example, if the data of the temperature sensor is missing at a certain moment, linear interpolation can be performed based on the temperature values at the previous and subsequent moments to estimate the missing value.

[0034] Removing outliers: Outliers may be data points that deviate significantly from the normal range due to sensor failures, human errors, or other unexpected factors. Set thresholds or use statistical methods (such as the 3σ principle) to identify and remove outliers. For example, for the harmonic characteristic data of motor current, if a data point exceeds several times the normal operating current and does not conform to the operating law of the equipment, it can be regarded as an outlier and removed.

[0035] Frequency band division and energy calculation:

[0036] Frequency band division: According to the vibration characteristics of the internal parts of the grinding machine and the fault characteristic frequencies, divide the frequency range of the collected vibration signals into different frequency bands. For example, the low-frequency band (such as 0 - 100Hz) can be used to analyze the overall operating state and basic vibration of the grinding machine, the middle-frequency band (100 - 1000Hz) can be used to observe the vibration characteristics of components such as the gearbox, and the high-frequency band (above 1000Hz) can be used to detect the fault characteristic frequencies of high-speed rotating components such as bearings. Usually, the fast Fourier transform (FFT) is used to convert the vibration signals in the time domain into frequency domain signals, and then the frequency bands are divided as needed.

[0037] Energy calculation: Calculate the energy of the signals in each frequency band. The energy can be obtained by integrating or summing the squared amplitudes of the signals in the frequency band. The energy value can be used as a characteristic parameter to reflect the vibration intensity in different frequency bands. For example, a sudden increase in the energy in a certain frequency band may indicate abnormal wear or failure of the components corresponding to that frequency band.

[0038] Temperature feature extraction:

[0039] Statistical features: Calculate statistical features such as the mean, variance, maximum value, and minimum value of the temperature signal. The mean can reflect the average temperature level of the parts, the variance can reflect the temperature fluctuation, and the maximum and minimum values can help judge whether there are extreme situations of too high or too low temperature. For example, if the mean of the bearing temperature continues to rise and exceeds the normal operating temperature range, it may indicate abnormal friction or lubrication problems with the bearing.

[0040] Change trend features: Analyze the change trend of temperature over time and extract features such as the rising slope and the falling slope. If the rising slope of the temperature is too large, it means that the temperature of the parts rises too fast, which may have potential fault hazards. Time series analysis methods such as the autoregressive moving average model (ARMA) can also be used to model and predict the change trend of temperature in order to detect abnormal temperature changes in a timely manner.

[0041] Thermal image features (combined with infrared thermal imaging sensor data): For the spatial temperature distribution data monitored by the infrared thermal imaging sensor, features of the thermal image can be extracted, such as the position, size, and temperature contrast of the hot spot area. The hot spot area may correspond to the faulty part of the internal component. By analyzing the changes in the hot spot area, the development trend of the fault can be judged.

[0042] Motor current harmonic feature extraction:

[0043] Harmonic component analysis: Perform Fourier transform on the motor current signal to obtain its spectrum and analyze the harmonic components therein. Under normal conditions, the harmonic content of the motor current is relatively low and conforms to a certain pattern. When problems such as overload, rotor imbalance, or stator winding fault occur in the motor, it will cause changes in the current harmonic components. For example, rotor imbalance may cause an increase in the amplitude of specific frequency harmonics. By monitoring information such as the amplitude, frequency, and phase of these characteristic harmonics, the operating state of the motor can be judged.

[0044] Characteristic parameter extraction: Extract some characteristic parameters from the current harmonic spectrum, such as the total harmonic distortion rate (THD), harmonic amplitude ratio (such as the ratio of the amplitude of a certain harmonic to the fundamental wave amplitude), etc. THD can reflect the overall harmonic content in the current signal, and the harmonic amplitude ratio can highlight the changes in specific harmonics. These characteristic parameters can be used as important bases for judging motor faults.

[0045] Lubricating oil metal particle content feature extraction:

[0046] Particle counting and size distribution: Analyze the data collected by the inductive particle sensor to count the number of metal particles in the lubricating oil and the distribution of particles in different size ranges. Generally speaking, as the wear of components such as bearings intensifies, the number of metal particles in the lubricating oil will increase, and the proportion of large-size particles may rise. For example, when it is found that the number of small-size particles gradually increases and large-size particles occasionally appear, it may indicate that the slight wear of the bearing is developing; while if a large number of large-size particles appear, it may indicate that the bearing has suffered relatively serious wear.

[0047] Particle concentration change trend: Analyze the change trend of the metal particle content over time and observe whether the particle concentration gradually increases, remains stable, or suddenly changes. A continuous increase in the particle concentration is usually a signal of increasing wear of equipment components, and a sudden large change may be related to sudden faults of the equipment or abnormalities in the lubricating oil system. Time series analysis methods such as the exponential smoothing method can be used to smooth and predict the change trend of the particle concentration to more accurately judge the wear state of the equipment.

[0048] As a further solution of the present invention, the data cleaning process includes: using a moving average filtering algorithm to remove sensor noise, and the specific process is as follows: where x(t), y(t), and z(t) respectively represent the original acceleration values on the x-axis, y-axis, and z-axis at time t, respectively represent the filtered acceleration values at time t, N represents the moving window size, which indicates the number of data points used to calculate the mean, and k represents the index variable within the moving window, indicating the offset of the current data point relative to time t; in order to more comprehensively reflect the overall vibration state of the device, calculate the sum of the three-axis acceleration vectors at each time point t to obtain the total vibration signal: Through the above steps, the moving average filtering algorithm can effectively remove periodic noise and high-frequency noise in the sensor measurement data, while retaining the overall trend of the signal, making the subsequent analysis of the internal part state of the grinding machine more accurate and reliable.

[0049] As a further solution of the present invention, the specific process of frequency band division and energy calculation is as follows: Frequency domain characteristics: FFT spectrum analysis, perform FFT analysis on the vibration signal, and extract the energy of key frequency bands: Low frequency band (0 - 500Hz): corresponding to the energy of gear meshing frequency and low-frequency vibration characteristics; Medium frequency band (500 - 2000Hz): corresponding to the bearing fault characteristic frequency; High frequency band (2000 - 5000Hz): capture impact signals, and a sudden increase in high-frequency vibration energy is usually related to microscopic material failure. The total energy is the sum of the energies of the three frequency bands. The temperature characteristic extraction process is: collect temperature, calculate the temperature change rate, perform moving average filtering to eliminate noise, record local extrema, and calculate the temperature change rate according to a 10s time window.

[0050] Further, frequency band division and energy calculation

[0051] FFT spectrum analysis: Use the fast Fourier transform (FFT) to convert the vibration signal in the time domain to the frequency domain. Through FFT, the distribution of the frequency components of the vibration signal in the frequency spectrum diagram can be obtained. The abscissa represents the frequency of the vibration signal, and the ordinate represents the frequency amplitude.

[0052] Frequency band division: According to the fault characteristics and vibration characteristics of the internal parts of the grinding machine, the entire frequency range is divided into three key frequency bands. The low frequency band (0 - 500Hz) corresponds to the energy of gear meshing frequency and low-frequency vibration characteristics, the medium frequency band (500 - 2000Hz) corresponds to the bearing fault characteristic frequency, and the high frequency band (2000 - 5000Hz) is used to capture impact signals. A sudden increase in high-frequency vibration energy is usually related to microscopic material failure.

[0053] Energy calculation: Calculate the energy in each frequency band separately. For discrete spectrum data, the energy of a frequency band can be approximately calculated by the sum of the squared amplitudes of all frequency points within that band. That is, for each frequency band, square the amplitudes of each frequency point within the band and then sum them up to obtain the energy of that band. The total energy is the sum of the energies of the three frequency bands.

[0054] Temperature feature extraction

[0055] Collect temperature: Through temperature sensors installed at parts such as the bearings and gearboxes of the grinding machine, collect the current temperature signals of the parts in real time.

[0056] Calculate the temperature change rate: Calculate the temperature change rate according to a 10s time window, that is, calculate the ratio of the change in temperature within two adjacent 10s time windows to the time interval, so as to reflect how fast the temperature changes with time.

[0057] Moving average filtering: Perform moving average filtering on the collected temperature data and the calculated temperature change rate data. Use the moving average filtering algorithm to remove sensor noise, make the temperature data smoother, and eliminate the interference of noise.

[0058] Record local extrema: In the filtered temperature data, record the local extreme points of the temperature, including local maxima and local minima. These extreme points may reflect some key features of the temperature change of the internal parts of the grinding machine.

[0059] As a further solution of the present invention, the process of extracting the harmonic characteristics of the motor current is as follows: Filtering and denoising: Adopt the moving average filtering algorithm, and the formula is as follows: where I(t) is the original current signal, is the filtered signal, N is the size of the moving window, Harmonic decomposition: Perform a fast Fourier transform on the filtered current signal to extract the amplitudes or energies of the fundamental wave and each harmonic; Total harmonic distortion rate calculation: The total harmonic distortion rate reflects the distortion degree of the current waveform, and the formula is: where I1 is the amplitude of the fundamental wave, I his the amplitude of the h - th harmonic. The energy proportion of the characteristic frequency band: For a specific harmonic frequency band, calculate the energy proportion. Filtering and denoising: The sliding mean filtering algorithm is adopted. The principle of this algorithm is to calculate the mean value of the original current signal within a sliding window of a fixed size, so as to smooth the signal and remove noise. For the original current signal I(t), the filtered signal I'(t) is obtained through this algorithm. The specific calculation method is to sum all the signal values within the sliding window centered on the current time t and divide by the window size N, and take the obtained average value as the filtered signal value at time t. This can effectively reduce the random noise in the signal and make the subsequent analysis results more accurate. Harmonic decomposition: Perform a fast Fourier transform (FFT) on the current signal after filtering and denoising. FFT is an efficient algorithm used to convert a time - domain signal into a frequency - domain signal. Through FFT, the current signal can be decomposed into the superposition of sine - wave components of different frequencies, so as to obtain the amplitude or energy information of the fundamental wave and each harmonic. In the frequency domain, the fundamental - wave component corresponds to the operating frequency during the normal operation of the motor, while each harmonic component reflects various abnormal conditions that may exist during the operation of the motor, such as motor winding faults, load imbalance, etc. By extracting the amplitude or energy of these harmonic components, it can provide an important basis for motor fault diagnosis. The energy proportion of the characteristic frequency band: Calculate the energy proportion for a specific harmonic frequency band. Different harmonic frequency bands may be related to different fault modes or operating states of the motor. For example, some low - frequency harmonics may be related to the load change or mechanical faults of the motor, while high - frequency harmonics may be related to the insulation damage of the motor winding or electromagnetic interference. By calculating the energy proportion of a specific harmonic frequency band, it is possible to more specifically analyze the operating state of the motor, highlight the characteristic information related to faults, and improve the accuracy and reliability of fault diagnosis.

[0060] As a further solution of the present invention, in the process of extracting the characteristics of the metal particle content in the lubricating oil, the sliding mean filtering algorithm is used to eliminate the sensor noise: where C(t) is the original concentration signal, is the filtered signal, and N is the size of the sliding window; the average particle concentration C debris :

[0061] As a further solution of the present invention, in step S3, the vibration spectrum energy, the temperature change rate, the motor current harmonic characteristics, and the lubricating oil metal particle concentration are used as input features to construct a joint feature vector: where E low (0 - 500Hz): reflects the gear meshing frequency and the overall vibration of the equipment, E mid (500 - 2000Hz): corresponds to the bearing fault characteristic frequency (such as outer - ring, inner - ring faults), E high(2000 - 5000Hz): Capture the instantaneous impact of crack propagation or material spalling, temperature change rate The temperature change rate within a 10 - second window, the concentration of lubricating oil metal particles (C debris ): Quantify the degree of bearing wear, the current harmonic distortion rate (THD): Reflect abnormal motor load. By combining these characteristic parameters into a joint feature vector, the operating state of the grinder can be comprehensively reflected from multiple angles, providing strong support for equipment fault diagnosis, predictive maintenance, etc. For example, in machine learning or data analysis, the joint feature vector can be used as input data to train a model to identify different equipment operating states or fault modes, thereby realizing intelligent monitoring and management of the grinder.

[0062] As a further solution of the present invention, specifically included in step S4 is to use a time - series model to capture the dynamic dependence relationship between vibration and temperature change. The calculation of the temperature residual predicted at the i - th time point:

[0063] ε i = |T 实际 (i) - T 预测 (i)|, the mean value (μ ε ): Represents the average value of all temperature residuals within the sliding window, characterizing the central position of the residuals; Among them, N represents the size of the sliding window, ε i represents the temperature residual predicted at the i - th time point, the standard deviation (σ ε ): The degree of dispersion of the residual data, measuring the fluctuation range of the residuals;

[0064] Among them, N represents the size of the sliding window, ε i represents the temperature residual predicted at the i - th time point, μ ε represents the mean value, dynamic threshold setting: Set an adaptive threshold based on the historical residual distribution, covering 99.7% of the normal data fluctuations:

[0065] Threshold = μ ε + 3σ ε , where μ ε : Residual mean value, σ ε : Standard deviation; Early warning trigger conditions: The residuals continuously exceed the threshold for more than 3 consecutive time windows; The temperature change rate exceeds 80% of the historical maximum value. Through the above steps, the time - series model can effectively capture the dynamic dependence relationship between vibration and temperature change, timely discover abnormal situations during equipment operation, and provide strong support for equipment fault diagnosis and maintenance.

[0066] The present invention has the following beneficial effects:

[0067] 1. Precisely capture the vibration state of the device: The sliding mean filtering algorithm is used to eliminate sensor noise, enabling more accurate acquisition of the original signal. By performing FFT spectrum analysis to extract the energy in different frequency bands, such as the low-frequency band reflecting gear meshing and the overall vibration of the device, the middle-frequency band corresponding to the characteristic frequencies of bearing faults, and the high-frequency band capturing impact signals, the vibration state of the device can be comprehensively reflected, which helps to detect potential problems of device parts in advance, such as gear wear, bearing faults, crack propagation, etc.

[0068] 2. Effectively monitor temperature changes: Collect the temperature and calculate the temperature change rate. After eliminating noise through sliding mean filtering, record the local extreme values, and calculate the temperature change rate according to the time window, which can sensitively reflect the heat generation and transfer inside the device, promptly detect abnormal heating, and provide important clues for fault diagnosis.

[0069] 3. Accurately analyze the harmonics of the motor current: The sliding mean filtering algorithm is used to filter and denoise the motor current signal. By performing fast Fourier transform to extract the amplitudes or energies of the fundamental wave and each harmonic, and calculating the harmonic distortion rate and the proportion of the energy in specific harmonic frequency bands, the operating state of the motor can be effectively monitored, and electrical problems such as load imbalance and winding faults can be detected in a timely manner.

[0070] 4. Quantify the concentration of metal particles in the lubricating oil: Use the sliding mean filtering algorithm to eliminate sensor noise and calculate the average value of the particle concentration, which can quantify the degree of bearing wear and indirectly understand the wear condition of internal parts of the device, providing a basis for device maintenance.

[0071] 5. Construct a combined feature vector for comprehensive evaluation: Use the vibration spectrum energy, temperature change rate, motor current harmonic characteristics, and lubricating oil metal particle concentration as input features to construct a combined feature vector, which comprehensively reflects the operating state of the device from multiple perspectives and provides strong support for device fault diagnosis, predictive maintenance, etc.

[0072] 6. Timely warn of device abnormalities: Use a time series model to capture the dynamic dependence relationship between vibration and temperature changes, calculate the temperature residuals and their mean and standard deviation, set a dynamic threshold based on the historical residual distribution. When the residuals continuously exceed the threshold or the temperature change rate exceeds 80% of the historical maximum value, a warning is triggered, which can promptly detect abnormal situations during the operation of the device, gain time for device maintenance, and prevent the further deterioration of faults.

[0073] To more clearly illustrate the structural features and functions of the present invention, the following will specifically describe the present invention in detail in combination with the accompanying drawings and specific embodiments. Description of the Drawings

[0074] Figure 1 It is the flowchart of the model establishment mentioned in the present invention. Detailed Embodiments

[0075] The present invention will be further described below in conjunction with the accompanying drawings and relevant knowledge. It will be clearly and completely described. Obviously, the described applications are only a part of the embodiments of the present invention, rather than all embodiments.

[0076] Embodiment 1. Refer to Figure 1 As shown, a method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis includes the following steps:

[0077] Step S1: Collect grinding data through sensor deployment. The sensors are temperature sensors, vibration sensors, infrared thermal imaging sensors, magnetoresistive current sensors, and inductive particle sensors. The specific data collection process is as follows: Install temperature sensors on the bearings and gearboxes of the grinding machine to collect the current temperature signals of the parts, with a sampling frequency ≥ 1 Hz; install vibration sensors to capture the high-frequency vibration signals of the parts, with a sampling frequency ≥ 5 kHz; install infrared thermal imaging sensors to monitor the heat generation of internal parts and achieve spatial temperature distribution monitoring; monitor the harmonic characteristics of the motor current through magnetoresistive current sensors to identify overload or rotor imbalance; measure the metal particle content in the lubricating oil through inductive particle sensors and judge the bearing wear degree in combination with the vibration signal;

[0078] Step S2: Preprocess and extract features from the collected data. Data preprocessing includes data cleaning, frequency band division and energy calculation, temperature feature extraction, motor current harmonic feature extraction, and lubricating oil metal particle content feature extraction. The data cleaning process includes: Using the sliding mean filtering algorithm to eliminate sensor noise. The specific process is as follows: where x(t), y(t), and z(t) respectively represent the original acceleration values on the x-axis, y-axis, and z-axis at time t, respectively represent the filtered acceleration values at time t, N represents the sliding window size, which is the number of data points used to calculate the mean, and k represents the index variable within the sliding window, which represents the offset of the current data point relative to time t; To more comprehensively reflect the overall vibration state of the device, calculate the sum of the three-axis acceleration vectors at each time point t to obtain the total vibration signal: The specific processes of frequency band division and energy calculation are as follows: Frequency domain characteristics: FFT spectrum analysis, perform FFT analysis on the vibration signal, and extract the energy of key frequency bands: Low frequency band (0 - 500 Hz): corresponding to the energy of gear meshing frequency and low frequency vibration characteristics; Medium frequency band (500 - 2000 Hz): corresponding to the bearing fault characteristic frequency; High frequency band (2000 - 5000 Hz): capture impact signals, and the sudden increase in high frequency vibration energy is usually related to microscopic material failure. The total energy is the sum of the energies of the three frequency bands. The temperature characteristic extraction process is as follows: collect temperature, calculate the temperature change rate, perform moving average filtering to eliminate noise, record local extrema, and calculate the temperature change rate according to a 10 - s time window; The motor current harmonic characteristic extraction process is as follows: Filtering and denoising: Use the moving average filtering algorithm, and the formula is as follows: where I(t) is the original current signal, is the filtered signal, N is the moving window size. Harmonic decomposition: Perform fast Fourier transform on the filtered current signal to extract the amplitudes or energies of the fundamental wave and each harmonic; Harmonic distortion rate calculation: The total harmonic distortion rate reflects the distortion degree of the current waveform, and the formula is: where I1 is the fundamental wave amplitude, I h is the amplitude of the h - th harmonic. Energy ratio of characteristic frequency bands: Calculate the energy ratio for a specific harmonic frequency band; The lubricating oil metal particle content characteristic extraction process is as follows: Use the moving average filtering algorithm to eliminate sensor noise: where C(t) is the original concentration signal, is the filtered signal, N is the moving window size; The average particle concentration C debris :

[0079] Step S3: Based on the data of pre - processing and feature extraction, perform model construction and training. Use the vibration spectrum energy, temperature change rate, motor current harmonic characteristics, and lubricating oil metal particle concentration as input features to construct a joint feature vector: where E low (0 - 500 Hz): reflects the gear meshing frequency and the overall vibration of the equipment, E mid (500 - 2000 Hz): corresponds to the bearing fault characteristic frequency (such as outer ring and inner ring faults), E high (2000 - 5000 Hz): captures the instantaneous impact of crack propagation or material spalling, the temperature change rate is the temperature change rate within a 10 - s window, the lubricating oil metal particle concentration (C debris ): quantifies the bearing wear degree, the current harmonic distortion rate (THD): reflects the abnormal motor load;

[0080] Step S4: Anomaly detection and early warning are implemented based on the constructed model. The time series model is used to capture the dynamic dependency between vibration and temperature changes. The predicted temperature residual at the i-th time point is calculated as:

[0081] ε i =|T 实际 (i)-T 预测 (i)|, mean (μ ε ): represents the average value of all temperature residuals in the sliding window, representing the center position of the residuals; Where N represents the size of the sliding window, ε i represents the predicted temperature residual at the i-th time point, the standard deviation (σ ε ): The degree of dispersion of residual data, which measures the fluctuation range of residual;

[0082] Where N represents the size of the sliding window, ε i represents the predicted temperature residual at the i-th time point, μ ε Represents the mean, dynamic threshold setting: Set adaptive thresholds based on historical residual distribution, covering 99.7% of normal data fluctuations:

[0083] Threshold = μ ε +3σ ε , where μ ε : Residual mean, σ ε : Standard deviation; Warning trigger conditions: The residual continues to exceed the threshold for more than 3 consecutive time windows; The temperature change rate exceeds 80% of the historical maximum value.

[0084] It should be noted in the present invention that: the present invention integrates multiple sensors for comprehensive monitoring: using multiple sensors such as temperature sensors, vibration sensors, infrared thermal imaging sensors, magnetoresistive current sensors and inductive particle sensors, it is possible to collect the operating data of the grinder from multiple dimensions, including part temperature, vibration signals, motor current harmonic characteristics, and metal particle content in the lubricating oil, etc., comprehensively reflecting the working status of the grinder, avoiding the limitations of single sensor monitoring, and improving the accuracy and reliability of detection.

[0085] Data preprocessing to improve accuracy: Data cleaning is performed through the sliding mean filter algorithm, which effectively eliminates sensor noise and improves data quality. Frequency band division and energy calculation of vibration signals can extract key frequency band energy related to gear meshing, bearing failure, and material micro-failure, providing a strong feature basis for fault diagnosis. At the same time, operations such as temperature feature extraction, motor current harmonic feature extraction, and lubricant metal particle content feature extraction further explore useful information in the data and improve the analysis accuracy of the internal parts status of the grinder.

[0086] Combined Feature Vector, Comprehensive Evaluation: The vibration spectrum energy, temperature change rate, motor current harmonic characteristics, and lubricating oil metal particle concentration are used as input features to construct a combined feature vector, comprehensively considering the influence of multiple factors on the abnormal temperature of the internal parts of the grinding machine. It can more comprehensively and accurately describe the operating state of the grinding machine, providing richer information for anomaly detection and early warning.

[0087] Time Series Model Analysis, Dynamic Monitoring: A time series model is used to capture the dynamic dependence relationship between vibration and temperature changes, enabling the timely discovery of potential connections and abnormal changes between vibration and temperature. By calculating the mean and standard deviation of the temperature residuals and setting dynamic thresholds, temperature anomalies can be detected more sensitively, and the thresholds can be adaptively adjusted according to the historical residual distribution, improving the accuracy and adaptability of anomaly detection. The setting of early warning trigger conditions, such as the residual continuously exceeding the threshold for more than 3 consecutive time windows and the temperature change rate exceeding 80% of the historical maximum value, can send early warning signals in a timely manner to remind operators to take corresponding measures to avoid the occurrence and further deterioration of equipment failures.

[0088] Intelligent Detection, Efficiency Improvement: This detection method realizes the automatic detection and early warning of abnormal temperatures of the internal parts of the grinding machine, reducing the influence of manual intervention and subjective judgment, and improving the detection efficiency and reliability. At the same time, based on data-driven model construction and training, the detection algorithm can be continuously optimized to adapt to the operating state of the grinding machine under different working conditions, providing strong support for the intelligent management and maintenance of grinding equipment, helping to improve production efficiency, reduce equipment failure rates, and maintenance costs.

[0089] In the present invention, regarding the relationship between various parameters and faults: Vibration spectrum energy: When mechanical components such as bearings and gears of the grinding machine exhibit early wear, cracks, and other faults, the energy in the frequency band corresponding to the fault characteristic frequency of the vibration signal will increase significantly. For example, faults such as inner ring faults of bearings and gear meshing problems have their respective specific frequency ranges. By monitoring the energy changes in these frequency bands, potential faults can be detected in a timely manner. Temperature change rate: When parts generate temperature changes due to friction or thermal effects, the rate of temperature rise is closely related to the change in vibration energy. When mechanical components develop faults, the vibration energy increases and the friction intensifies, resulting in a faster temperature rise. Therefore, the temperature change rate can be used as an indirect indicator of the development of faults to assist in judging the working state of the parts. Motor current harmonic characteristics: Motor current harmonics are caused by internal electrical asymmetry or load changes. Rotor imbalance will increase specific harmonic components in the current, thereby generating additional electromagnetic forces and causing mechanical vibrations. Therefore, monitoring the motor current harmonic characteristics can identify problems such as motor overload or rotor imbalance, indirectly reflecting the operating conditions of the internal parts of the grinding machine. Lubricating oil metal particle concentration: By detecting the content of metal particles in the lubricating oil using an oil analysis sensor, the wear conditions of components such as bearings can be understood. When bearings are worn, metal particles will fall off into the lubricating oil. Combining with the vibration signal, the wear degree of the bearings can be judged more accurately, providing an important basis for fault diagnosis.

[0090] Establishment of the data fusion model: Based on the relationships between the above-mentioned various parameters and faults, the relevance modeling is used to establish a data fusion model by capturing the dynamic relevance of the vibration signal spectrum energy, lubricating oil metal particle concentration, motor current harmonic characteristics, and temperature change rate. Specifically, it is to comprehensively consider the changes of these parameters over time and their mutual relationships. For example, when the vibration spectrum energy suddenly increases in certain specific frequency bands, it may be accompanied by an acceleration of the temperature change rate, a change in the motor current harmonic characteristics, and an increase in the lubricating oil metal particle concentration. By analyzing the time-series relationships of these parameters and using methods such as machine learning and data analysis, a model that can accurately reflect the working state of the internal parts of the grinding machine is constructed, thereby realizing the early prediction of abnormal part temperatures.

[0091] Joint prediction logic: In practical applications, by monitoring the time-series relationships of sudden increases in vibration spectrum energy, changes in motor current harmonic characteristics, increases in lubricating oil metal particle concentration, and accelerations in the temperature change rate, abnormal working conditions of the parts can be identified in advance. When these parameters show abnormal changes, the model will analyze and judge according to preset rules and algorithms, and issue corresponding warning signals to remind the operator to take timely measures, such as further checking the equipment, performing maintenance or replacing parts, etc., to avoid the occurrence and expansion of equipment failures and ensure the normal operation of the grinding machine.

[0092] Infrared thermal imaging detection monitors the spatial temperature distribution of the heat generated by internal parts during the monitoring process, helping to accurately locate overheated parts and providing more precise guidance for maintenance. Together with other parameter monitoring methods, it constitutes a complete state monitoring and fault warning system for the grinding machine.

[0093] Embodiment 2: A method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis, comprising the following steps:

[0094] Step 1: Sensor deployment and data synchronization

[0095] Temperature sensor: Install thermocouples at key parts such as the bearings and gearboxes of the grinding machine to collect the current temperature signals of the parts, with a sampling frequency ≥ 1 Hz.

[0096] Vibration sensor: Install a three-axis acceleration sensor to capture the high-frequency vibration signals of the parts, with a sampling frequency ≥ 5 kHz.

[0097] Infrared thermal imaging sensor: Install an infrared thermal imaging sensor to monitor the heat generation of internal parts and achieve spatial temperature distribution monitoring.

[0098] Magnetoresistive current sensor: Monitor the harmonic characteristics of the motor current to identify overload or rotor imbalance problems.

[0099] Inductive particle sensor: Measure the metal particle content in the lubricating oil and combine the vibration signals to judge the bearing wear degree.

[0100] Step 2: Data preprocessing and feature extraction

[0101] Data cleaning: Use the sliding mean filtering algorithm to remove sensor noise.

[0102]

[0103] Where x(t), y(t), and z(t) respectively represent the original acceleration values (including noise) on the x-axis, y-axis, and z-axis at time t. respectively represent the filtered acceleration values (after denoising) at time t. N represents the size of the sliding window, indicating the number of data points used to calculate the mean. The larger N is, the stronger the smoothing effect, but the greater the signal delay. The smaller N is, the opposite is true. k represents the index variable within the sliding window, indicating the offset of the current data point relative to time t.

[0104] To more comprehensively reflect the overall vibration state of the equipment, calculate the vector sum of the three-axis accelerations at each time point t to obtain the total vibration signal:

[0105]

[0106] Frequency band division and energy calculation:

[0107] Frequency domain features: FFT spectrum analysis (extracting the energy of the bearing fault characteristic frequency band).

[0108] Perform FFT analysis on the vibration signal and extract the energy of the key frequency bands:

[0109] Table 1, low frequency band (0 - 500 Hz): mainly corresponding to the energy of the gear meshing frequency and low frequency vibration characteristics. The gear meshing frequency is usually determined by the rotational speed and the number of teeth and is common in the low frequency range. For example, when the rotational speed is 1000 RPM, the meshing frequency is

[0110] If the number of teeth is 30, then f 啮合 = 500 Hz.

[0111] Table 1

[0112]

[0113] Table 2, middle frequency band (500 - 2000 Hz): corresponding to the bearing fault characteristic frequency. Typical value range: the fault frequency of common industrial bearings (such as 6205) is usually located in 500 - 2000 Hz. For example, BPFO ≈ 100 - 300 Hz, BPFI ≈ 200 - 600 Hz, BSF ≈ 500 - 1500 Hz.

[0114] For example, the calculation formula for the bearing outer race fault frequency (BPFO) is:

[0115]

[0116] The calculation formula for the bearing inner race fault frequency (BPFI) is:

[0117]

[0118] The calculation formula for the rolling element fault frequency (BSF) is:

[0119]

[0120] Among them, n is the number of bearing rolling elements, d is the diameter of the rolling element (mm), D is the pitch diameter of the bearing (mm), θ is the contact angle, and f r is the shaft rotation frequency (Hz).

[0121] Table 2

[0122]

[0123] Table 3, high frequency band (2000 - 5000 Hz): capturing impact signals (such as instantaneous events like crack propagation and local spalling). A sudden increase in high frequency vibration energy is usually related to microscopic material failure.

[0124] Table 3

[0125] Sub - frequency band Frequency range Corresponding fault type Minor impact 2000 - 3000Hz Micro - crack propagation, local pitting Medium impact 3000 - 4000Hz Material spalling, local fracture Severe impact 4000 - 5000Hz Sudden fracture, severe crack propagation

[0126] For the synthesized total vibration signal a total Perform a fast Fourier transform (FFT) to obtain the complex spectrum A total (f): A total (f) = FFT(a total (t))

[0127] Spectrum amplitude calculation:

[0128]

[0129] Calculate the energy proportion of each frequency band:

[0130]

[0131] Where X(f) is the spectrum amplitude at frequency f, f start and f end are the start and end frequencies of the frequency band respectively, E band is the frequency band energy, the low-frequency energy is E low the middle-frequency energy is E mid and the high-frequency energy is E high .

[0132] The total energy is the sum of the energies of the three frequency bands:

[0133] E total = E low + E mid + E high

[0134] Calculation of the energy proportion of each frequency band:

[0135]

[0136] Temperature feature extraction: Collect temperature (sampling frequency 1 Hz), calculate the temperature change rate Perform a moving average filter (window length 10 s) to eliminate noise and record local extrema (such as the highest temperature change rate per hour).

[0137] Calculate the temperature change rate according to a 10 s time window:

[0138]

[0139] Motor current harmonic feature extraction:

[0140] Filtering and denoising: Adopt a moving average filter algorithm (the same as the vibration signal), and the formula is as follows:

[0141]

[0142] Among them, I(t) is the original current signal, is the filtered signal, and N is the sliding window size.

[0143] Harmonic decomposition: Perform a fast Fourier transform (FFT) on the filtered current signal to extract the amplitudes or energies of the fundamental wave (50 / 60 Hz) and each harmonic (such as the 2nd, 3rd, and 5th harmonics).

[0144] Calculation of harmonic distortion rate: The total harmonic distortion rate (THD) reflects the degree of distortion of the current waveform, and the formula is:

[0145]

[0146] Among them, I1 is the amplitude of the fundamental wave, and I h is the amplitude of the hth harmonic.

[0147] Energy proportion in characteristic frequency bands: Calculate the energy proportion for specific harmonic frequency bands (such as the 2x frequency corresponding to rotor imbalance and the sidebands related to bearing faults).

[0148] Content of metal particles in lubricating oil:

[0149] Data preprocessing

[0150] Use the sliding mean filtering algorithm to eliminate sensor noise:

[0151]

[0152] Among them, C(t) is the original concentration signal, is the filtered signal, and N is the sliding window size.

[0153] Average particle concentration C debris :

[0154]

[0155] Step 3: Model construction and training

[0156] Use the vibration spectrum energy (low frequency, medium frequency, high frequency), temperature change rate, motor current harmonic characteristics, and lubricating oil metal particle concentration as input features to construct a joint feature vector:

[0157]

[0158] Among them, E low (0 - 500 Hz): Reflects the gear meshing frequency and the overall vibration of the equipment

[0159] E mid(500 - 2000 Hz): Corresponds to the bearing fault characteristic frequencies (such as outer race, inner race faults).

[0160] E high (2000 - 5000 Hz): Captures the instantaneous impact of crack propagation or material spalling.

[0161] Rate of temperature change Rate of temperature change within a 10 - second window.

[0162] Concentration of lubricating oil metal particles (C debris ): Quantifies the degree of bearing wear.

[0163] Total harmonic distortion (THD) of current: Reflects abnormal motor load (such as rotor imbalance).

[0164] Step 4: Abnormality detection and warning

[0165] Adopts a time - series model (LSTM) to capture the dynamic dependence relationship between vibration and temperature changes.

[0166] Model input: Organizes the combined feature vector F into a sliding window input according to the time series. Sliding window: For example, a 60 - second window, each window contains 60 time steps, and each time step inputs 6 - dimensional features.

[0167] Model processing process:

[0168] Forget gate: Determines which historical information to discard. For example: If the current vibration energy (E mid ) suddenly increases, the forget gate may retain the vibration features of the previous time step and filter out irrelevant current harmonic fluctuations.

[0169] Input gate: Updates the cell state and records the association between the current features and temperature. For example: If the current significantly increases, the input gate will strengthen the association between vibration energy and temperature change.

[0170] Cell state: Stores long - term dependence information. For example: The continuously accumulating concentration of metal particles (C debris ) will gradually change the cell state, reflecting the wear trend.

[0171] Output gate: Generates a hidden state based on the current cell state for predicting the temperature residual. For example: If the cell state records an event of sudden increase in vibration energy, the output gate will generate a signal related to the temperature residual.

[0172] Calculation of the predicted temperature residual at the i - th time point: ε i =∣T 实际 (i)-T 预测 (i)∣

[0173] Mean (με ) represents the average value of all temperature residuals within the sliding window, characterizing the central position of the residuals.

[0174]

[0175] where N represents the size of the sliding window, and ε i represents the predicted temperature residual at the i-th time point.

[0176] Standard deviation (σ ε ) represents the degree of dispersion of the residual data, measuring the fluctuation range of the residuals.

[0177]

[0178] where N represents the size of the sliding window, and ε i represents the predicted temperature residual at the i-th time point, and μ ε represents the mean value.

[0179] Dynamic threshold setting: An adaptive threshold is set based on the historical residual distribution (normal distribution) to cover 99.7% of the normal data fluctuations:

[0180] Threshold = μ ε + 3σ ε

[0181] where μ ε : Residual mean value, and σ ε : Standard deviation

[0182] Early warning trigger conditions: 1. The residual continuously exceeds the threshold for more than 3 consecutive time windows (30s); 2. The temperature change rate exceeds 80% of the historical maximum value.

[0183] In the present invention, through sensor deployment and data synchronization: A variety of sensors are used to comprehensively collect data, covering temperature, vibration, infrared thermal imaging, motor current harmonics, and lubricating oil metal particle concentration, etc., providing rich information for subsequent analysis.

[0184] Data preprocessing and feature extraction: Data cleaning: The sliding mean filtering algorithm is used to remove sensor noise, and the smoothing effect and signal delay are adjusted according to the window size N. Frequency band division and energy calculation: The vibration signal is analyzed by FFT, divided into three frequency bands: low, medium, and high, corresponding to different fault types respectively, and the energy and proportion of each frequency band are calculated. Temperature feature extraction: Temperature is collected, the change rate is calculated, and noise is eliminated by sliding mean filtering and local extreme values are recorded. Motor current harmonic feature extraction: After filtering and denoising, FFT analysis is carried out, the fundamental wave and the amplitudes or energies of each harmonic are extracted, and the harmonic distortion rate and the proportion of energy in the characteristic frequency band are calculated. Lubricating oil metal particle content: The sliding mean filtering is used to eliminate noise, and the average value of particle concentration is calculated.

[0185] Model construction and training: Construct a joint feature vector from vibration spectrum energy, temperature change rate, motor current harmonic characteristics, and lubricating oil metal particle concentration as the model input.

[0186] Anomaly detection and warning: Use the LSTM time series model to capture the dynamic dependence between vibration and temperature changes. Calculate the mean and standard deviation of the temperature residuals, and set a dynamic threshold based on the historical residual distribution. Trigger a warning when the residuals continuously exceed the threshold or the temperature change rate exceeds 80% of the historical maximum value.

[0187] Advantage analysis of the present invention, multi-source data fusion: Integrate multiple sensor data to reflect the operating state of the grinder from multiple dimensions, improving the accuracy and reliability of fault detection. Fine feature extraction: Conduct a detailed frequency band division and energy calculation on the vibration signal to accurately identify different types of faults. Dynamic threshold setting: Set an adaptive threshold based on the historical residual distribution, which can adapt to different working conditions and equipment state changes. Application of time series model: Use the LSTM model to capture the dynamic correlation between vibration and temperature, and achieve early prediction of temperature anomalies.

[0188] The technical principle of the present invention is described above in combination with specific embodiments, which are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. Those skilled in the art can readily think of other specific embodiments of the present invention without creative efforts, and these embodiments will fall within the protection scope of the present invention.

Claims

1. A method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis, characterized in that, It includes the following steps: Step S1: Collect the operating data of the grinding machine by deploying a variety of sensors, including a temperature sensor, a three-axis vibration sensor, an infrared thermal imaging sensor, a magnetoresistive current sensor, and an inductive particle sensor; Step S2: Preprocess and extract features from the collected data; Step S3: Build and train a model based on the preprocessed and feature-extracted data; Step S4: Implement anomaly detection and early warning according to the built model.

2. The method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis according to claim 1, characterized in that In the step S1, the specific data collection process is as follows: The current temperature signal of the parts is collected by installing temperature sensors on the bearings and gearboxes of the grinding machine, and the sampling frequency ≥ 1Hz; By installing a three-axis vibration sensor, the high-frequency vibration signal of the parts is captured, and the sampling frequency ≥ 5kHz; By installing an infrared thermal imaging sensor, the heat generation of the internal parts is monitored to realize the monitoring of the spatial temperature distribution; The harmonic characteristics of the motor current are monitored by a magnetoresistive current sensor to identify overload or rotor imbalance states. The metal particle content in the lubricating oil is measured by an inductive particle sensor, and the bearing wear degree is evaluated in combination with the vibration signal.

3. The method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis according to claim 1, characterized in that, In the step S2, data preprocessing includes data cleaning: The sensor noise is removed by using a sliding mean filtering algorithm; Band division and energy calculation: The vibration signal is subjected to FFT spectrum analysis to extract the energy characteristics of the low-frequency band (0 - 500Hz), the middle-frequency band (500 - 2000Hz), and the high-frequency band (2000 - 5000Hz); Temperature Feature extraction: Calculate the temperature change rate and perform sliding mean filtering to eliminate noise; Motor current harmonic feature extraction: After filtering and denoising, the harmonic amplitude and total harmonic distortion rate are extracted by fast Fourier transform; Lubricating oil metal particle content feature extraction: The noise is eliminated by using sliding mean filtering, and the average particle concentration is calculated.

4. The method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis according to claim 3, characterized in that, The data cleaning process includes: using a moving average filtering algorithm to eliminate sensor noise. The specific process is as follows: where x(t), y(t), and z(t) respectively represent the original acceleration values on the x-axis, y-axis, and z-axis at time t, respectively represent the filtered acceleration values at time t, N represents the moving window size, indicating the number of data points used to calculate the mean, and k represents the index variable within the moving window, indicating the offset of the current data point relative to time t; to more comprehensively reflect the overall vibration state of the device, calculate the sum of the three-axis acceleration vectors at each time point t to obtain the total vibration signal:

5. The method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis according to claim 4, wherein, The specific process of the band division and energy calculation is as follows: Frequency domain feature: FFT spectrum analysis, the vibration signal is subjected to FFT analysis to extract the energy of the key frequency bands: Low-frequency band (0 - 500Hz): Corresponding to the energy of the gear meshing frequency and the low-frequency vibration characteristics; Middle-frequency band (500 - 2000Hz): Corresponding to the bearing fault characteristic frequency; High-frequency band (2000 - 5000Hz): Capture the impact signal. The sudden increase in high-frequency vibration energy is usually related to the microscopic failure of the material. The total energy is the sum of the energies of the three frequency bands. The temperature feature extraction process is: Collect the temperature, calculate the temperature change rate, perform sliding mean filtering to eliminate noise, record the local extreme values, and calculate the temperature change rate according to a 10s time window.

6. The method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis according to claim 5, wherein The process of extracting the harmonic characteristics of the motor current is as follows: Filtering and denoising: The sliding mean filtering algorithm is adopted, and the formula is as follows: where I(t) is the original current signal, is the filtered signal, N is the size of the sliding window, Harmonic decomposition: Perform a fast Fourier transform on the filtered current signal to extract the amplitudes or energies of the fundamental wave and each harmonic; Calculation of harmonic distortion rate: The total harmonic distortion rate reflects the distortion degree of the current waveform, and the formula is: where I1 is the amplitude of the fundamental wave, I h is the amplitude of the h-th harmonic, Proportion of energy in the characteristic frequency band: Calculate the proportion of energy for a specific harmonic frequency band.

7. The temperature anomaly detection method for internal parts of a grinding machine based on vibration signal analysis according to claim 6, characterized in that The process of extracting the characteristics of the metal particle content in the lubricating oil is as follows: The sliding mean filtering algorithm is used to eliminate sensor noise: where C(t) is the original concentration signal, is the filtered signal, N is the size of the sliding window; the mean value of the particle concentration C debris :

8. The method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis according to claim 7, characterized in that, In the step S3, the vibration spectrum energy, the temperature change rate, the motor current harmonic characteristics, and the lubricating oil metal particle concentration are used as input features to construct a joint feature vector: where, E low (0 - 500 Hz): reflects the gear meshing frequency and the overall vibration of the equipment, E mid (500 - 2000 Hz): corresponds to the bearing fault characteristic frequencies (such as outer ring, inner ring faults), E high (2000 - 5000 Hz): captures the instantaneous impact of crack propagation or material spalling, the temperature change rate The temperature change rate within a 10 - second window, the lubricating oil metal particle concentration (C debris ): quantifies the bearing wear degree, the current harmonic distortion rate (THD): reflects the abnormal motor load.

9. The method for detecting abnormal temperature of internal parts of a grinding machine based on vibration signal analysis according to claim 8, characterized in that, In the step S4, it specifically includes using a time series model to capture the dynamic dependence relationship between vibration and temperature changes, and calculating the predicted temperature residual at the i-th time point: ε i = |T 实际 (i) - T 预测 (i)|, mean (μ ε ): represents the average value of all temperature residuals within the sliding window, characterizing the central position of the residuals; Among them, N represents the size of the sliding window, ε i represents the predicted temperature residual at the i-th time point, standard deviation (σ ε ): the degree of dispersion of the residual data, measuring the fluctuation range of the residuals; where N represents the size of the sliding window, and ε i represents the predicted temperature residual at the i-th time point, and μ ε represents the mean value. Dynamic threshold setting: An adaptive threshold is set based on the historical residual distribution to cover 99.7% of the normal data fluctuations: Threshold = μ ε + 3σ ε , where μ ε : mean of residuals, σ ε : standard deviation; Early warning trigger conditions: The residuals continuously exceed the threshold for more than 3 consecutive time windows; The temperature change rate exceeds 80% of the historical maximum value.

Images