A method and system for fault diagnosis of oil monitoring based on the Internet of Things

Abstract

This invention relates to the field of fault diagnosis technology, specifically to an IoT-based oil monitoring fault diagnosis method and system, comprising the following steps: collecting mechanical vibration signals through oil monitoring, setting a time window, performing a short-time Fourier transform to convert the time-domain signal into a frequency-domain signal, and obtaining spectrum identification data. In this invention, IoT technology enables real-time processing of batch data and training of a classifier using machine learning methods, further improving the accuracy of fault type identification. The introduction of entropy calculation allows for the assessment of the machine's operating status and efficiency. By comparing with set fault level standards, mechanical fault risks can be promptly identified and prioritized. This combination of real-time monitoring and automatic identification greatly improves preventative maintenance capabilities and equipment operating efficiency, reduces the risk of unexpected downtime, and brings significant benefits in cost control and productivity improvement.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a method and system for fault diagnosis of oil monitoring based on the Internet of Things. Background Technology

[0002] The field of fault diagnosis technology focuses on identifying potential problems by detecting anomalies in equipment or systems. This involves collecting and analyzing various types of data, including physical parameters such as temperature, pressure, and vibration, to determine whether the equipment is functioning correctly or requires maintenance. Fault diagnosis methods can be based on traditional sensor technology or integrate modern information technologies such as the Internet of Things (IoT), big data, and machine learning to improve the accuracy and efficiency of diagnosis. The application of these technologies can significantly reduce downtime, improve production efficiency, and extend equipment lifespan.

[0003] Among them, the IoT-based oil monitoring and fault diagnosis method refers to using IoT technology to monitor and analyze the state of oil in mechanical equipment, such as oil temperature, viscosity, and impurity content. By monitoring changes in oil in real time, faults and wear of mechanical equipment can be detected promptly. The main applications of this technology include extending the service life of mechanical equipment, preventing sudden failures, and optimizing maintenance plans to achieve cost-effectiveness and operational efficiency in industrial production. Summary of the Invention

[0004] To address the significant shortcomings of existing technologies that rely on traditional sensors and periodic monitoring, particularly in terms of real-time performance and fault prevention, this invention provides a method and system for fault diagnosis of oil monitoring based on the Internet of Things (IoT). The system cannot promptly reflect the dynamic changes in mechanical equipment during operation, delaying fault detection and leading to equipment damage or production line downtime, thus impacting overall production efficiency and maintenance costs. Traditional technologies for oil monitoring cannot reflect minute changes in oil quality in real time; if these changes are not detected promptly, they can cause premature wear or failure of mechanical components. In production environments that heavily rely on efficiency and stability, this deficiency results in significant economic losses and safety risks.

[0005] On the one hand, an IoT-based oil monitoring fault diagnosis method is provided, which includes:

[0006] S1: Collect mechanical vibration signals through oil monitoring, set a time window, perform short-time Fourier transform to convert the time domain signal into a frequency domain signal, and obtain spectrum identification data;

[0007] S2: Based on the spectrum identification data, analyze the energy spectral density, accumulate the energy of the data in the target frequency range, identify the frequency characteristics associated with mechanical faults, and obtain the fault indication frequency;

[0008] S3: Using the fault indication frequency, by calculating the degree of change in the differentiated time series, abnormal modes that deviate from the normal vibration mode are identified, and abnormal vibration modes are obtained.

[0009] S4: Using the vibration anomaly pattern as input data, the machine learning classifier is trained using the Internet of Things. By setting target parameters, the fault type in the vibration data is identified, and the fault type identification result is obtained.

[0010] S5: Using the fault type identification results, extract fault samples from the oil monitoring data, perform multiple stages of entropy calculation, calculate the data entropy at different time points, record the entropy data, evaluate the operating speed and efficiency of the machinery, and obtain the entropy data record.

[0011] S6: Based on the entropy data records, compare them with the pre-set fault level standards, identify the health status of the machinery by threshold judgment, and sort the processing priority of mechanical fault risks to obtain fault risk assessment results.

[0012] As a further aspect of the present invention, the spectrum identification data includes the key frequencies, corresponding amplitudes, and spectrum distribution of the vibration signal; the fault indication frequency includes the key fault frequency, associated energy peak value, and frequency bandwidth; the vibration anomaly mode includes the abnormal frequency mode, deviation degree, and mode stability; the fault type identification result includes the identified fault type, fault severity, and confidence score; the entropy data record includes the calculated entropy value, entropy change rate, and time series entropy value comparison; and the fault risk assessment result includes the assessed fault level, risk ranking, and expected maintenance time point.

[0013] As a further aspect of the present invention, the steps of collecting mechanical vibration signals through oil monitoring, setting a time window, performing a short-time Fourier transform to convert the time-domain signal into a frequency-domain signal, and obtaining spectrum identification data are as follows:

[0014] S101: Collect mechanical vibration signals through oil monitoring, set time window parameters, segment the vibration signals, check the signal quality by adjusting the signal segmentation, and perform normalization processing to adjust the amplitude of the segmented signals to generate segmented time domain signals.

[0015] S102: Based on the segmented time-domain signal, short-time Fourier transform technology is used to set the frequency resolution during the transformation process. By dynamically adjusting the transformation window length, the frequency domain characteristics of the signal in the differentiated time window are calculated to obtain frequency domain signal data.

[0016] S103: Using the frequency domain signal data, perform spectrum identification, filter frequency components by peak detection, and identify key spectral features by combining signal strength and frequency position to obtain spectrum identification data.

[0017] As a further aspect of the present invention, the formula for the short-time Fourier transform technique is as follows:

[0018]

[0019] Where X′(ω) is the frequency domain signal data, ω is the angular frequency, t is the time offset, n is the discrete time index, x(n) is the signal amplitude at time point n, w(nt) is a window function with a width of w centered at t, α is the time window weight attenuation coefficient, β is the frequency window weight attenuation coefficient, and e is the natural constant.

[0020] As a further aspect of the present invention, the steps of analyzing the energy spectral density based on the spectrum identification data, accumulating the energy of data within the target frequency range, identifying frequency characteristics associated with mechanical faults, and obtaining the fault indication frequency are as follows:

[0021] S201: Based on the spectrum identification data, perform energy spectral density analysis, organize the energy of each frequency component, evaluate the energy contribution of each frequency point through signal strength, merge the frequency point energies, and obtain the cumulative energy spectrum;

[0022] S202: Define a target frequency range using the accumulated energy spectrum, iteratively aggregate and analyze the accumulated energy data within the target range, identify frequency regions with energy anomalies by setting thresholds, and generate target frequency analysis results;

[0023] S203: Using the target frequency analysis results, the identified abnormal frequency features are compared with the known mechanical fault frequency features. Through matching analysis, frequency features associated with mechanical faults are identified, and fault indication frequencies are generated.

[0024] As a further aspect of the present invention, the specific steps for identifying abnormal vibration modes that deviate from normal vibration modes by using the fault indication frequency and calculating the degree of change in the differentiated time series are as follows:

[0025] S301: Based on the fault indication frequency, identify the frequency difference between the current vibration data of the mechanical equipment and the standard vibration mode, and obtain the frequency difference time series by analyzing the degree of change of each fault indication frequency through differential time series analysis;

[0026] S302: Using the frequency difference time series, perform time window analysis on the data in the series, and distinguish between normal and abnormal vibration modes by calculating the degree of deviation of the vibration mode in each time window, and generate time window analysis results;

[0027] S303: Using the analysis results of the time window, the identified abnormal vibration patterns are compared with normal vibration patterns. Vibration behavior that deviates from the normal pattern is verified by pattern recognition technology, and an abnormal vibration pattern is generated.

[0028] As a further aspect of the present invention, the steps of using the vibration anomaly pattern as input data, training a machine learning classifier using the Internet of Things, identifying fault types in the vibration data by setting target parameters, and obtaining fault type identification results are as follows:

[0029] S401: Using the vibration anomaly pattern as input data, collect and transmit data using IoT devices, initialize and set parameters for the machine learning classifier, train the classifier, and obtain a training dataset.

[0030] S402: Using the training dataset, perform the classifier training process, adjust the target parameters in the classification, including the learning rate and the number of iterations, optimize the response sensitivity to vibration anomaly patterns, and generate parameter training results;

[0031] S403: Based on the training results of the parameters, the classifier is applied to the current vibration data. By analyzing the data, the fault type is identified, and the prediction result is matched with the known fault type to generate the fault type identification result.

[0032] As a further aspect of the present invention, the steps of extracting fault samples from oil monitoring data using the fault type identification results, performing multiple stages of entropy calculation, calculating data entropy at differentiated time points, recording entropy data, and evaluating the operating speed and efficiency of the machinery to obtain entropy data recording are as follows:

[0033] S501: Based on the fault type identification result, extract the target fault sample from the oil monitoring data corresponding to the fault type, extract the key data points from the target fault sample, and obtain the selected fault sample data.

[0034] S502: Using the selected fault sample data, perform entropy calculation, evaluate the randomness and volatility of the data points, calculate the data entropy at different time points, and generate volatility evaluation results;

[0035] S503: Based on the volatility assessment results, analyze the overall data entropy trend, and by recording and analyzing the entropy value changes, assess the speed and efficiency changes of mechanical operation to obtain entropy value data records.

[0036] As a further aspect of the present invention, the steps of comparing the entropy data records with a pre-set fault level standard, identifying the health status of the machinery through threshold determination, and prioritizing the handling of mechanical fault risks to obtain fault risk assessment results are as follows:

[0037] S601: Based on the entropy data record, a decision tree algorithm is used to compare and analyze multiple operating parameters and entropy values ​​of the machine, make an initial judgment on the fault level, and obtain the fault level judgment result;

[0038] The formula for the decision tree algorithm is as follows:

[0039]

[0040] Where R is the fault level judgment result, x a Let t be the actual entropy value of the a-th operating parameter of the machine. a w is the preset fault entropy threshold value for the a-th operating parameter. a Let d be the weighting coefficient of the a-th running parameter. a λ is the parameter deviation adjustment coefficient, N is the number of operating parameters, and e is the natural constant.

[0041] S602: Using the fault level judgment result, perform threshold judgment, conduct risk assessment on the differentiated fault levels, and rank the severity of mechanical faults by analyzing the risk coefficient and the probability of fault occurrence, and generate fault risk ranking result;

[0042] S603: Based on the fault risk ranking results, integrate and compare the fault risk data of different levels, assess the health status of the machinery through oil monitoring, identify and record the handling priority of mechanical fault risks, and obtain the fault risk assessment results.

[0043] On the other hand, an IoT-based oil monitoring fault diagnosis system is provided, which is applied to an IoT-based oil monitoring fault diagnosis method. The system includes:

[0044] The signal acquisition module collects mechanical vibration signals through oil monitoring, sets the acquisition time window, performs a short-time Fourier transform on the acquired time-domain signal, converts it into a frequency-domain signal, and obtains spectrum data.

[0045] The energy analysis module uses the spectrum data to analyze the energy spectral density, performs energy accumulation on the data within the selected target frequency range, and identifies the fault indication frequency;

[0046] The abnormal pattern recognition module obtains the vibration abnormal pattern by analyzing the changes in the differential time series based on the fault indication frequency.

[0047] The machine learning training module inputs the vibration anomaly pattern into the machine learning classifier, trains the classifier to identify the fault type in the differential vibration data, and generates fault type identification results.

[0048] The entropy analysis module uses the fault type identification results to extract fault samples from the oil monitoring data. By analyzing the data entropy at different time points, it records the entropy data, compares it with the set fault level standards, identifies the health status of the machinery, and obtains the fault risk assessment results.

[0049] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0050] By acquiring and analyzing mechanical vibration signals, the accuracy and real-time performance of fault diagnosis are significantly improved. By performing a short-time Fourier transform, the time-domain signal is effectively converted into a frequency-domain signal, and energy spectral density analysis of the frequency-domain signal enables precise identification of frequency characteristics associated with mechanical faults. Utilizing IoT technology, batch data can be processed in real time, and a classifier can be trained using machine learning methods, further improving the accuracy of fault type identification. The introduction of entropy calculation allows for the assessment of the machine's operating status and efficiency; by comparing it with predefined fault level standards, mechanical fault risks can be promptly identified and prioritized. This combination of real-time monitoring and automatic identification greatly enhances preventative maintenance capabilities and equipment operating efficiency, reduces the risk of unexpected downtime, and brings significant benefits in cost control and productivity improvement. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the workflow of the present invention;

[0052] Figure 2 This is a detailed flowchart of S1 of the present invention;

[0053] Figure 3 This is a detailed flowchart of the S2 process of the present invention;

[0054] Figure 4 This is a detailed flowchart of the S3 process of the present invention;

[0055] Figure 5 This is a detailed flowchart of the S4 process of the present invention;

[0056] Figure 6 This is a detailed flowchart of S5 of the present invention;

[0057] Figure 7 This is a detailed flowchart of S6 of the present invention;

[0058] Figure 8 This is a system flowchart of the present invention. Detailed Implementation

[0059] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0060] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0061] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0062] Please see Figure 1 This invention provides an IoT-based oil monitoring fault diagnosis method, the processing flow of which may include the following steps:

[0063] S1: Collect mechanical vibration signals through oil monitoring, set a time window, perform short-time Fourier transform to convert the time domain signal into a frequency domain signal, and initialize and organize the spectrum data to obtain spectrum identification data;

[0064] S2: Based on spectrum identification data, analyze energy spectral density, accumulate energy in data within the target frequency range, identify frequency characteristics associated with mechanical faults, and obtain fault indication frequency;

[0065] S3: By using the fault indication frequency and calculating the degree of change in the differentiated time series, abnormal modes that deviate from the normal vibration mode are identified, and the abnormal vibration mode is obtained.

[0066] S4: Using the vibration anomaly pattern as input data, the machine learning classifier is trained using the Internet of Things. By setting target parameters, the fault type in the vibration data is identified, and the fault type identification result is obtained.

[0067] S5: Utilize the fault type identification results to extract fault samples from the oil monitoring data, perform multiple stages of entropy calculation, calculate the data entropy at differentiated time points, record entropy data, evaluate the operating speed and efficiency of the machinery, and obtain entropy data records;

[0068] S6: Based on entropy data records, compare them with pre-set fault level standards, identify the health status of the machinery through threshold judgment, prioritize the handling of mechanical fault risks, and obtain fault risk assessment results.

[0069] The spectrum identification data includes the key frequencies, corresponding amplitudes, and spectral distribution of the vibration signal; the fault indication frequencies include the key fault frequencies, associated energy peaks, and frequency bandwidth; the vibration anomaly modes include the abnormal frequency modes, deviation levels, and mode stability; the fault type identification results include the identified fault type, fault severity, and confidence score; the entropy data records include the calculated entropy value, entropy change rate, and time series entropy value comparison; and the fault risk assessment results include the assessed fault level, risk ranking, and expected maintenance time.

[0070] Please see Figure 2 The specific steps for collecting mechanical vibration signals through oil monitoring, setting a time window, performing a short-time Fourier transform to convert the time-domain signal into a frequency-domain signal, and obtaining spectrum identification data are as follows:

[0071] S101: The execution flow of collecting mechanical vibration signals through oil monitoring, setting time window parameters, segmenting the vibration signals, checking the signal quality by adjusting the signal segmentation, and adjusting the amplitude of the segmented signals through normalization processing to generate segmented time domain signals is as follows.

[0072] In mechanical vibration signal monitoring, to ensure signal quality, the signal is segmented by setting a time window parameter. The core of this process lies in determining the optimal length of the time window, which needs to be adjusted according to the mechanical operating characteristics and vibration characteristics. For example, for mechanical equipment with periodic rapid changes, the time window should be shorter to capture detailed vibration information. Conversely, for equipment with stable changes, a longer time window can be set to reduce the amount of data processing. Signal normalization processing includes adjusting the signal amplitude to a certain range to avoid data distortion caused by excessively large or small amplitudes in subsequent analysis. For example, the amplitude can be normalized to between -1 and 1 to ensure the comparability of data collected from different equipment or at different times. The signal after amplitude adjustment is easier for subsequent processing and analysis, ensuring data quality and analysis accuracy, and generating segmented time-domain signals.

[0073] S102: Based on the segmented time-domain signal, the short-time Fourier transform technique is adopted. The frequency resolution in the transformation process is set. By dynamically adjusting the transformation window length, the frequency domain characteristics of the signal in the differentiated time window are calculated, and the frequency domain signal data is obtained. The execution flow is as follows:

[0074] The formula for the short-time Fourier transform technique is as follows:

[0075]

[0076] Where X′(ω) is the frequency domain signal data, ω is the angular frequency, t is the time offset, n is the discrete time index, x(n) is the signal amplitude at time point n, w(nt) is a window function with a width of w centered at t, α is the time window weight attenuation coefficient, β is the frequency window weight attenuation coefficient, and e is the natural constant.

[0077] The definitions and calculation processes of the parameters involved in the formula are as follows:

[0078] ω (angular frequency): This is used in frequency domain analysis to represent the frequency of each component in a signal. It is determined by the sampling rate of the signal. If the sampling rate of the signal is set to 8000Hz, then the angular frequency ω can range from 0 to 4000Hz.

[0079] t (time offset): represents the center point of the window function in the signal. For example, if the sampling frequency is 8000Hz, a specific window function is selected with the center at the 1000th sample point of the signal.

[0080] n (Discrete Time Index): Represents the length range of the entire signal, the sample index from the beginning to the end of the signal. For example, for a signal with a length of 1 second at a sampling rate of 8000Hz, n ranges from 0 to 7999.

[0081] x(n) (signal amplitude): The actual amplitude of the signal at time point n, which is obtained by reading the actual acquired signal data;

[0082] w(nt) (window function): Uses Hanning or Hamming window, with window width w set to 256 sample points, meaning each window covers 256 sample points of signal;

[0083] α (Time window weight decay coefficient): Controls the decay rate at the edge of the window function, determined through experiments or optimization algorithms. For the Hanning window, the empirical value of α is set to 0.46.

[0084] β (Frequency window weight attenuation coefficient): Adjusts the accuracy of frequency resolution. For audio signals, this value is obtained by minimizing spectral leakage in overlapping areas, and a commonly used value is 0.001.

[0085] Example of calculation derivation: Set t = 1000, ω = 2π × 1000 (corresponding to 1000Hz), α = 0.46, β = 0.001, and use the Hanning window;

[0086] Calculate w(100-n) and x(n):

[0087]

[0088] calculate and

[0089]

[0090] Calculate the partial values ​​of X′(ω) in combination:

[0091]

[0092] Because of e -372600 The summation value approaches 0, indicating that under extremely high attenuation conditions, almost all signal energy is lost. This result demonstrates that a high attenuation coefficient leads to complete suppression of signal energy at this frequency, highlighting the need for careful parameter selection when designing window functions and attenuation coefficients to ensure the effectiveness and accuracy of signal analysis.

[0093] S103: Using frequency domain signal data, perform spectrum identification, filter frequency components through peak detection, and identify key spectral features by combining signal strength and frequency position. The execution process for obtaining spectrum identification data is as follows:

[0094] In the processing of frequency domain signal data, spectrum identification is a crucial step. Peak detection is used to filter frequency components. This process is accomplished by identifying significant peaks in the spectrum. Each peak represents a major frequency component. For example, by setting a threshold, only frequency components with amplitudes exceeding that threshold are considered. Combining signal strength and frequency location, key spectral features can be identified, including the exact location of the frequency and the corresponding amplitude value. This data is essential for subsequent mechanical condition monitoring, helping to identify potential faults or anomalies and obtain spectrum identification data.

[0095] Please see Figure 3 The specific steps for obtaining the fault indication frequency by analyzing energy spectral density based on spectrum identification data, accumulating energy in data within the target frequency range, identifying frequency characteristics associated with mechanical faults, and obtaining the fault indication frequency are as follows:

[0096] S201: Based on the spectrum identification data, perform energy spectral density analysis, organize the energy of each frequency component, evaluate the energy contribution of each frequency point through signal strength, merge the frequency point energy, and obtain the cumulative energy spectrum. The execution flow is as follows:

[0097] In the processing of spectral data, it is necessary to evaluate the energy contribution of each frequency point. This requires obtaining the energy of each frequency component based on energy spectral density analysis, calculating the energy value of each frequency point through mathematical transformation, and accumulating the energy of all frequency points. This process involves the organization of spectral data and energy calculation, ensuring the accuracy and feasibility of the analysis and providing basic data for subsequent steps. This analysis not only helps to understand the energy distribution of the signal at different frequencies, but also provides key information for subsequent fault analysis, resulting in the accumulated energy spectrum.

[0098] S202: The execution flow is as follows: by accumulating energy spectrum, defining target frequency range, iteratively gathering and analyzing accumulated energy data within the target range, identifying frequency regions with energy anomalies by setting thresholds, and generating target frequency analysis results;

[0099] By setting a threshold, frequency regions of energy anomalies can be identified, according to formula E. total =∑E z Calculate the accumulated energy over the entire target frequency range. Where E total E represents the total accumulated energy within the target frequency range. z This represents the energy at a single frequency point z. Determine the energy E at each frequency point z. z The total accumulated energy E can be obtained directly from frequency analysis equipment or calculated using spectrum data processing algorithms. For example, if five frequency points are set with energy values ​​of 10, 20, 30, 40, and 50 respectively, then the total accumulated energy E is... total =10+20+30+40+50=150. The key is to perform a simple arithmetic summation of the energy at all target frequency points to obtain the total energy. The total value is the basis for assessing whether there is an energy anomaly in the frequency region.

[0100] S203: Using the target frequency analysis results, the identified abnormal frequency features are compared with the known mechanical fault frequency features. The frequency features associated with mechanical faults are identified through matching analysis, and the execution flow for generating fault indication frequencies is as follows;

[0101] When identifying frequency characteristics associated with mechanical faults, the target frequency analysis results are compared with known mechanical fault frequency characteristics. The comparison process includes a detailed analysis of each abnormal frequency characteristic to ensure accurate matching of frequency characteristics. This comparison is processed by algorithms, such as using least squares or statistical matching techniques, to identify frequency characteristics associated with mechanical faults. This process ensures the accuracy of the analysis and the reliability of the operation. Comparative analysis is not only a key step in identifying faults but also a necessary measure to ensure the safe operation of equipment, generating fault indication frequencies.

[0102] Please see Figure 4 The specific steps for identifying abnormal vibration modes by using fault indication frequencies and calculating the degree of change in differentiated time series are as follows:

[0103] S301: Based on the fault indication frequency, identify the frequency difference between the current vibration data of the mechanical equipment and the standard vibration mode. Analyze the degree of change of each fault indication frequency through differential time series analysis. The execution flow of the frequency difference time series is as follows.

[0104] Analyzing the frequency differences between current vibration data and standard vibration modes of mechanical equipment involves differential time series analysis. By evaluating the degree of change in the frequency of each fault indication, this analysis process ensures accurate monitoring and analysis of vibration data, helps to understand the vibration characteristics of mechanical equipment under different operating conditions, and provides data support for subsequent fault prevention and maintenance. By recording and accumulating each measurement difference between the fault frequency and the standard frequency, further analysis and fault diagnosis are conducted, generating a frequency difference time series.

[0105] S302: The execution flow of using frequency difference time series to perform time window analysis on the data in the series, and distinguishing between normal and abnormal vibration modes by calculating the degree of deviation of vibration modes within each time window, and generating time window analysis results is as follows;

[0106] By calculating the degree of deviation of the vibration mode within each time window, according to the formula... Calculate the deviation. In the formula, ΔP represents the degree of deviation of the vibration mode, and P... obs P represents the vibration mode observed in each time window. norm This represents the normal vibration mode. The vibration data P observed within each time window... obs and standard vibration data P norm To calculate the difference, for example, if the vibration data observed within a certain time window are 15, 20, and 22, while the normal data are 10, 15, and 20, then the sum of squares of the difference is calculated as (15-10). 2 +(20-15) 2 +(22-20) 2 =25+25+4=54, then take the square root to get the degree of deviation. The key to the calculation steps lies in quantifying the degree of vibration deviation through mathematical methods, providing numerical basis for identifying abnormal vibrations.

[0107] S303: Using the time window analysis results, the identified abnormal vibration patterns are compared with normal vibration patterns. The vibration behavior that deviates from the normal pattern is verified by pattern recognition technology. The execution flow for generating abnormal vibration patterns is as follows.

[0108] The process of comparing identified abnormal vibration patterns with normal vibration patterns is achieved through pattern recognition technology. During the comparative analysis, vibration behaviors that deviate from the normal pattern are verified. This analysis process not only helps to quickly identify potential equipment faults, but also effectively guides maintenance personnel to carry out targeted repairs. The identification of abnormal vibration patterns is achieved by calculating and comparing the deviation of vibration data in each time window with the standard vibration pattern, ensuring the accuracy and reliability of vibration analysis, providing important technical support for the health status monitoring of mechanical equipment, and generating abnormal vibration patterns.

[0109] Please see Figure 5 The specific steps for using abnormal vibration patterns as input data, training a machine learning classifier using the Internet of Things, identifying fault types in vibration data by setting target parameters, and obtaining fault type identification results are as follows:

[0110] S401: The execution flow is as follows: Based on the vibration anomaly pattern as input data, data is collected and transmitted using IoT devices, the machine learning classifier is initialized and its parameters are set, and the classifier is trained to obtain the training dataset.

[0111] Initialize and set parameters for the machine learning classifier according to the formula. Calculate the loss function. In the formula, L(θ) represents the loss function, θ represents the classifier parameters, and N... i y represents the total number of samples. i This represents the true label of the i-th sample. This represents the predicted label. Initialize the classifier parameters θ. For example, if the initial parameter value is 0, and the sample data is {0, 1, 0, 1}, the initial predicted label value is 0.5. Substituting these values ​​into the formula, the loss function is calculated as follows:

[0112] The key to this calculation step lies in adjusting the parameters by optimizing the loss function to improve the accuracy and robustness of the classifier.

[0113] S402: Using the training dataset, the classifier training process is executed, the target parameters in the classification are adjusted, including the learning rate and the number of iterations, the response sensitivity to vibration anomaly patterns is optimized, and the parameter training results are generated. The execution flow is as follows:

[0114] During the training of the classifier, the key lies in adjusting the target parameters in the classification, including the learning rate and the number of iterations. Optimizing the parameters can significantly improve the sensitivity to vibration anomaly patterns. Through mathematical optimization techniques, such as gradient descent, the parameters are adjusted to minimize the classification error. Through performance evaluation during the iteration process, it is ensured that each step of parameter adjustment is based on the learning results of the previous step. The optimization process includes multiple iterations, and each iteration evaluates the classifier performance and adjusts the parameters. By precisely controlling the training process, the optimal classifier settings are obtained, and parameter training results are generated.

[0115] S403: Based on the parameter training results, the classifier is applied to the current vibration data. By analyzing the data, the fault type is identified, and the prediction results are matched with known fault types to generate the fault type identification results. The execution flow is as follows:

[0116] By applying a classifier to the current vibration data and analyzing the data, specific fault types can be identified. In this process, the classifier is used based on previously trained parameters and models. The prediction results are matched with known fault types. This matching process is achieved through algorithm comparison. This process not only confirms the effectiveness of the classifier but also provides decision-making information for fault handling. The classifier responds quickly to and analyzes new vibration data to ensure the accuracy and efficiency of fault identification and generate fault type identification results.

[0117] Please see Figure 6 Using the fault type identification results, fault samples are extracted from the oil monitoring data. Multiple stages of entropy calculation are performed. By calculating the data entropy at differentiated time points, entropy data is recorded, and the operating speed and efficiency of the machinery are evaluated. The specific steps for recording entropy data are as follows:

[0118] S501: Based on the fault type identification result, extract the target fault sample from the oil monitoring data corresponding to the fault type, extract the key data points from the target fault sample, and obtain the selected fault sample data. The execution process is as follows:

[0119] Extracting target fault samples from oil monitoring data corresponding to the fault type is crucial. The key to this process is to accurately identify and extract specific sample data related to the identified fault type from batch monitoring data. Extracting key data points from the target fault samples is essential for in-depth analysis of the specific manifestations and causes of the fault. This involves data screening, feature labeling, and recording of key data points. These data points represent key monitoring indicators at the time of the fault occurrence, providing a foundation for subsequent data analysis and fault diagnosis. This ensures the relevance and accuracy of the analysis and yields selected fault sample data.

[0120] S502: The execution flow of using selected fault sample data to perform entropy calculation, evaluating the randomness and volatility of data points, calculating the data entropy at differentiated time points, and generating volatility assessment results is as follows;

[0121] By evaluating the randomness and volatility of the data points, the data entropy is calculated using the formula H = -∑p(x)logp(x). In this formula, H represents the data entropy, and p(x) represents the probability of data point x occurring. The probability p(x) of each data point is determined, and the data point distribution is set to {10, 10, 20, 20, 20, 30}. The probabilities of each point are then respectively... Substituting into the formula, the data entropy is calculated as follows: The key to this calculation step is to express the uncertainty and complexity of the data through entropy values, providing a numerical basis for volatility assessment.

[0122] S503: Based on the volatility assessment results, analyze the overall data entropy trend, and evaluate the speed and efficiency changes of mechanical operation by recording and analyzing the entropy value changes. The execution process for entropy value data recording is as follows;

[0123] Analyzing the overall data entropy trend is achieved through the continuous recording and analysis of entropy value changes in the data. The steps for assessing changes in the speed and efficiency of mechanical operation include entropy value calculation, trend plotting, and interpretation of changes. This not only reflects the stability and efficiency of the mechanical operating status but also provides a basis for predicting the future operating trend of the mechanical equipment. Entropy value analysis helps the maintenance team identify potential operational problems, ensuring the timeliness and effectiveness of mechanical maintenance and generating entropy value data records.

[0124] Please see Figure 7 The specific steps for obtaining a fault risk assessment result are as follows: Based on entropy value data records, the data is compared with pre-set fault level standards. Thresholds are used to determine the health status of the machinery, and the handling priority of mechanical fault risks is ranked.

[0125] S601: Based on entropy data records, the decision tree algorithm is used to compare and analyze multiple operating parameters and entropy values ​​of the machine to make an initial judgment on the fault level. The execution flow for obtaining the fault level judgment result is as follows;

[0126] The formula for the decision tree algorithm is as follows:

[0127]

[0128] Where R is the fault level judgment result, x a Let t be the actual entropy value of the a-th operating parameter of the machine. a w is the preset fault entropy threshold value for the a-th operating parameter. aLet d be the weighting coefficient of the a-th running parameter. a λ is the parameter deviation adjustment coefficient, N is the adjustment parameter, and e is the natural constant.

[0129] x a The actual entropy value is obtained in real time through the mechanical operation monitoring system. The machine is set with three parameters, and the monitored entropy values ​​are x1 = 0.2, x2 = 0.3, and x3 = 0.1.

[0130] t a : Preset fault entropy threshold values, set according to mechanical historical data and fault records, such as t1=0.5, t2=0.3, t3=0.2;

[0131] w a Weighting coefficients are set according to the degree of influence of the parameters. Here, w1 = 0.5, w2 = 0.3, and w3 = 0.2.

[0132] d a The parameter deviation adjustment coefficients are obtained from mechanical fault analysis data and are set as d1 = -1.0, d2 = 0, and d3 = 1.0.

[0133] N: Number of parameters, which is 3 in this example;

[0134] λ: Adjustment parameter used to adjust the denominator when the number of parameters is large; set to 0.5.

[0135] Calculate the adjustment value for each parameter.

[0136] For a = 1,

[0137] For a = 2,

[0138] For a = 3,

[0139] Multiply the adjusted values ​​by the weighting coefficients and sum them:

[0140]

[0141] Calculation results:

[0142] The results show that the calculated fault level judgment result is 0.029. This lower result indicates that the current mechanical state is close to the normal operating state. This reflects that the formula with the addition of adjustment coefficient and weight adjustment has increased the detail and accuracy of the calculation, making the fault level judgment more consistent with the actual operating conditions.

[0143] S602: Using the fault level judgment results, perform threshold judgment, conduct risk assessment on differentiated fault levels, and rank the severity of mechanical faults by analyzing risk coefficients and fault occurrence probabilities to generate fault risk ranking results. The execution process is as follows:

[0144] The threshold determination process involves risk assessment of differentiated fault levels, analyzing risk coefficients and fault occurrence probabilities, with a key focus on ranking the severity of mechanical faults. This analysis relies on mathematical models and statistical methods, such as calculating the cumulative distribution function or probability density function for each fault level to determine the fault risk coefficient. Ranking is then based on these coefficients, ensuring that the fault risk assessment is based on actual data and accurate algorithmic calculations. This provides scientific decision-making information for maintenance and prevention, generating a fault risk ranking result.

[0145] S603: Based on the failure risk ranking results, integrate and compare the failure risk data of different levels, assess the health status of the machinery through oil monitoring, identify and record the handling priority of mechanical failure risks, and obtain the failure risk assessment results. The execution process is as follows:

[0146] By monitoring and assessing the health of machinery through oil fluids, the priority of handling mechanical failure risks is identified and recorded. This process involves not only the integration and comparison of data, but also the monitoring and analysis of specific indicators in the mechanical oil. Through this comprehensive assessment, the types of failures that need to be prioritized can be effectively identified, ensuring the efficiency of maintenance work and the stable operation of mechanical equipment, and obtaining failure risk assessment results.

[0147] Please see Figure 8 On the other hand, an IoT-based oil monitoring fault diagnosis system is provided, which is applied to an IoT-based oil monitoring fault diagnosis method. The system includes:

[0148] The signal acquisition module collects mechanical vibration signals through oil monitoring, sets the acquisition time window, performs a short-time Fourier transform on the acquired time-domain signal, converts it into a frequency-domain signal, and obtains spectrum data.

[0149] The energy analysis module uses spectral data to analyze energy spectral density, performs energy accumulation on data within a selected target frequency range, and identifies fault indication frequencies.

[0150] The abnormal pattern recognition module obtains the vibration abnormal pattern by analyzing the changes in the differential time series based on the fault indication frequency.

[0151] The machine learning training module inputs the vibration anomaly pattern into the machine learning classifier, trains the classifier to identify the fault type in the differential vibration data, and generates fault type identification results.

[0152] The entropy analysis module uses the fault type identification results to extract fault samples from the oil monitoring data. By analyzing the data entropy at different time points, it records the entropy data, compares it with the set fault level standards, identifies the health status of the machinery, and obtains the fault risk assessment results.

[0153] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for fault diagnosis of oil monitoring based on the Internet of Things, characterized in that, The method includes: Mechanical vibration signals are collected by oil monitoring, a time window is set, and a short-time Fourier transform is performed to convert the time-domain signal into a frequency-domain signal to obtain spectrum identification data. Based on the spectrum identification data, the energy spectral density is analyzed, the data in the target frequency range are accumulated, the frequency characteristics associated with mechanical faults are identified, and the fault indication frequency is obtained. By using the fault indication frequency and calculating the degree of change in the differentiated time series, abnormal modes that deviate from the normal vibration mode are identified, and abnormal vibration modes are obtained. Using the vibration anomaly pattern as input data, the Internet of Things is used to train a machine learning classifier. By setting target parameters, the fault type in the vibration data is identified, and the fault type identification result is obtained. Using the fault type identification results, fault samples from oil monitoring data are extracted, and multiple stages of entropy calculation are performed. By calculating the data entropy at different time points, entropy data is recorded, and the operating speed and efficiency of the machinery are evaluated to obtain entropy data records. Based on the entropy data records, the data is compared with the pre-set fault level standards. The health status of the machinery is identified by threshold judgment, and the processing priority of the machinery fault risk is sorted to obtain the fault risk assessment result.

2. The method for fault diagnosis of oil monitoring based on the Internet of Things according to claim 1, characterized in that, The spectrum identification data includes the key frequencies, corresponding amplitudes, and spectral distribution of the vibration signal; the fault indication frequencies include the key fault frequencies, associated energy peaks, and frequency bandwidth; the vibration anomaly modes include the abnormal frequency modes, deviation levels, and mode stability; the fault type identification results include the identified fault type, fault severity, and reliability score; the entropy data records include the calculated entropy value, entropy change rate, and time series entropy value comparison; and the fault risk assessment results include the assessed fault level, risk ranking, and expected maintenance time.

3. The method for fault diagnosis of oil monitoring based on the Internet of Things according to claim 1, characterized in that, The specific steps for collecting mechanical vibration signals through oil monitoring, setting a time window, performing a short-time Fourier transform to convert the time-domain signal into a frequency-domain signal, and obtaining spectrum identification data are as follows: Mechanical vibration signals are collected by oil monitoring, time window parameters are set, vibration signals are segmented, signal quality is checked by adjusting signal segmentation, and the amplitude of the segmented signals is adjusted by standardization processing to generate segmented time domain signals. Based on the segmented time-domain signal, short-time Fourier transform technology is used to set the frequency resolution during the transformation process. By dynamically adjusting the transformation window length, the frequency domain characteristics of the signal in the differentiated time window are calculated to obtain frequency domain signal data. Using the frequency domain signal data, spectrum identification is performed. Frequency components are filtered by peak detection, and key spectral features are identified by combining signal strength and frequency position to obtain spectrum identification data.

4. The method for fault diagnosis of oil monitoring based on the Internet of Things according to claim 3, characterized in that, The formula for the short-time Fourier transform technique is as follows: Where X′(ω) is the frequency domain signal data, ω is the angular frequency, t is the time offset, n is the discrete time index, x(n) is the signal amplitude at time point n, w(nt) is a window function with a width of w centered at t, α is the time window weight attenuation coefficient, β is the frequency window weight attenuation coefficient, and e is the natural constant.

5. The method for fault diagnosis of oil monitoring based on the Internet of Things according to claim 1, characterized in that, Based on the spectrum identification data, the steps of analyzing the energy spectral density, accumulating the energy of data within the target frequency range, identifying frequency characteristics associated with mechanical faults, and obtaining the fault indication frequency are as follows: Based on the spectrum identification data, energy spectral density analysis is performed to organize the energy of each frequency component, evaluate the energy contribution of each frequency point through signal strength, and merge the frequency point energies to obtain the cumulative energy spectrum; The target frequency range is defined by the accumulated energy spectrum. The accumulated energy data within the target range is iteratively aggregated and analyzed. By setting a threshold, the frequency region with abnormal energy is identified, and the target frequency analysis result is generated. Using the target frequency analysis results, the identified abnormal frequency features are compared with known mechanical fault frequency features. Through matching analysis, frequency features associated with mechanical faults are identified, and fault indication frequencies are generated.

6. The method for fault diagnosis of oil monitoring based on the Internet of Things according to claim 1, characterized in that, The specific steps for identifying abnormal vibration modes by using the fault indication frequency and calculating the degree of change in the differentiated time series are as follows: Based on the fault indication frequency, the frequency difference between the current vibration data of the mechanical equipment and the standard vibration mode is identified, and the degree of change of each fault indication frequency is analyzed by differential time series analysis to obtain the frequency difference time series. Using the frequency difference time series, time window analysis is performed on the data in the series. By calculating the degree of deviation of the vibration mode in each time window, normal and abnormal vibration modes are distinguished, and time window analysis results are generated. Using the analysis results of the time window, the identified abnormal vibration patterns are compared with normal vibration patterns. Vibration behaviors that deviate from the normal patterns are verified by pattern recognition technology, and abnormal vibration patterns are generated.

7. The method for fault diagnosis of oil monitoring based on the Internet of Things according to claim 1, characterized in that, The specific steps for using the vibration anomaly pattern as input data, training a machine learning classifier using the Internet of Things, identifying fault types in the vibration data by setting target parameters, and obtaining fault type identification results are as follows: Using the vibration anomaly pattern as input data, data is collected and transmitted using IoT devices to initialize and set parameters for a machine learning classifier, and the classifier is trained to obtain a training dataset. Using the training dataset, the classifier training process is executed, the target parameters in the classification are adjusted, including the learning rate and the number of iterations, the response sensitivity to vibration anomaly patterns is optimized, and parameter training results are generated. Based on the training results of the parameters, the classifier is applied to the current vibration data. By analyzing the data, the fault type is identified, and the prediction results are matched with known fault types to generate fault type identification results.

8. The method for fault diagnosis of oil monitoring based on the Internet of Things according to claim 1, characterized in that, Using the fault type identification results, fault samples are extracted from the oil monitoring data. Multiple stages of entropy calculation are performed. By calculating the data entropy at differentiated time points, entropy data is recorded, and the operating speed and efficiency of the machinery are evaluated. The specific steps for recording entropy data are as follows: Based on the fault type identification results, target fault samples are extracted from the oil monitoring data corresponding to the fault type, key data points are extracted from the target fault samples, and selected fault sample data is obtained. Using the selected fault sample data, entropy calculation is performed. By evaluating the randomness and volatility of the data points, the data entropy at different time points is calculated, and volatility evaluation results are generated. Based on the volatility assessment results, the overall data entropy trend is analyzed. By recording and analyzing the changes in entropy values, the speed and efficiency changes of mechanical operation are assessed, and entropy data records are obtained.

9. The method for fault diagnosis of oil monitoring based on the Internet of Things according to claim 1, characterized in that, Based on the entropy data records, the following steps are taken to compare them with pre-set fault level standards, identify the health status of the machinery through threshold judgment, and prioritize the handling of mechanical fault risks to obtain fault risk assessment results: Based on the entropy data records, a decision tree algorithm is used to compare and analyze multiple operating parameters and entropy values ​​of the machine, make an initial judgment on the fault level, and obtain the fault level judgment result. The formula for the decision tree algorithm is as follows: Where R is the fault level judgment result, x a Let t be the actual entropy value of the a-th operating parameter of the machine. a w is the preset fault entropy threshold value for the a-th operating parameter. a Let d be the weighting coefficient of the a-th running parameter. a λ is the parameter deviation adjustment coefficient, N is the adjustment parameter, and e is the natural constant. Using the fault level judgment results, a threshold determination is performed to conduct a risk assessment on the differentiated fault levels. By analyzing the risk coefficient and the probability of fault occurrence, the severity of mechanical faults is ranked to generate a fault risk ranking result. By integrating and comparing the fault risk data of different levels based on the fault risk ranking results, assessing the health status of the machinery through oil monitoring, identifying and recording the handling priority of mechanical fault risks, and obtaining the fault risk assessment results.

10. An oil monitoring and fault diagnosis system based on the Internet of Things, characterized in that, The method for fault diagnosis of oil monitoring based on the Internet of Things according to any one of claims 1-9, wherein the system comprises: The signal acquisition module collects mechanical vibration signals through oil monitoring, sets the acquisition time window, performs a short-time Fourier transform on the acquired time-domain signal, converts it into a frequency-domain signal, and obtains spectrum data. The energy analysis module uses the spectrum data to analyze the energy spectral density, performs energy accumulation on the data within the selected target frequency range, and identifies the fault indication frequency; The abnormal pattern recognition module obtains the vibration abnormal pattern by analyzing the changes in the differential time series based on the fault indication frequency. The machine learning training module inputs the vibration anomaly pattern into the machine learning classifier, trains the classifier to identify the fault type in the differential vibration data, and generates fault type identification results. The entropy analysis module uses the fault type identification results to extract fault samples from the oil monitoring data. By analyzing the data entropy at different time points, it records the entropy data, compares it with the set fault level standards, identifies the health status of the machinery, and obtains the fault risk assessment results.

Images