Abnormity detection method and device for intermittent operation equipment

Abstract

The invention relates to the technical field of equipment anomaly detection, in particular to an anomaly detection method and device for intermittent operation equipment, and the method comprises the steps: obtaining the detection data of the intermittent operation equipment; identifying the detection data based on a multi-signal fusion operation state identification mechanism to determine an operation stage of the intermittent operation equipment, and processing the detection data in the operation stage based on a periodic alignment mode of a normalized time axis to obtain a state characteristic value of the intermittent operation equipment in each time interval of the operation stage; according to the state characteristic value in each time interval and a pre-created normal behavior baseline, determining an operation deviation degree condition of the intermittent operation equipment; and judging whether the intermittent operation equipment is abnormal or not according to the operation deviation degree condition. According to the method, under the condition that equipment is frequently stopped and started and sampling is interrupted, feature calculation and matching can be independently completed in each operation section, the stability and reliability of the detection process are ensured, and the anomaly detection precision is greatly improved.

Description

Technical Field

[0001] The present application relates to the technical field of equipment anomaly detection, and in particular to a method for detecting anomalies in intermittently operated equipment, an apparatus for detecting anomalies in intermittently operated equipment, a computer device, and a computer-readable storage medium. Background Art

[0002] Intermittently operating equipment is widely used in many fields, including modern industrial production and energy extraction. This type of equipment typically starts and stops periodically based on production needs, resulting in discontinuous data collected about the equipment's operation. For example, in oil and gas production, pumping units, as a core piece of equipment, need to operate intermittently based on well production; water injection pumps similarly start and stop periodically based on factors such as reservoir pressure. Currently, methods for detecting equipment anomalies are primarily based on the analysis of continuous time series data. This data processing approach often assumes that the data is continuous and stationary over time. By calculating statistical characteristics such as the mean, variance, and standard deviation of continuously collected data, anomalies can be detected.

[0003] Obviously, in related technologies, anomaly detection schemes based on the statistical feature analysis of continuous time series are highly dependent on data continuity. If zero current or a "flat line" signal appears during the equipment shutdown phase, the algorithm will lose its analytical basis or mistakenly identify a normal shutdown as a fault. At the same time, for intermittent equipment with short operating cycles and scarce samples, early fault characteristics are difficult to capture in a timely manner, resulting in a high rate of missed detection. Anomaly detection schemes based on interpolation / missing value processing, while performing linear interpolation, spline interpolation, or Gaussian process prediction to fill in missing data segments during the equipment shutdown period, restoring the data to a "continuous" sequence, and then applying traditional statistical analysis or classification models to determine anomalies, the "virtual data" generated by interpolation or prediction deviates significantly from the actual operating signal. This error is transmitted to subsequent detection links, similarly leading to false positives or missed detections. This results in low detection accuracy and poor versatility. The masking of breakpoint features also weakens sensitivity to intermittent characteristics. Summary of the Invention

[0004] (1) Technical issues to be resolved

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present application provides a method and device for detecting anomalies of intermittently running equipment, which solves the technical problems of the existing abnormality detection methods of intermittently running equipment, such as incomplete data, misjudgment of the stop phase, insufficient capture of short-cycle features, poor model adaptability and weak predictive maintenance capabilities.

[0006] (2) Technical solution

[0007] In order to achieve the above objectives, the main technical solutions adopted in this application include:

[0008] In a first aspect, an embodiment of the present application provides a method for detecting an anomaly of an intermittently operated device, comprising:

[0009] Acquiring detection data of the intermittently operated equipment, wherein the detection data is used to characterize the operating status of the intermittently operated equipment;

[0010] Identifying the detection data based on a multi-signal fusion operation state identification mechanism to determine the operation stage of the intermittent operation equipment, and processing the detection data of the operation stage based on a period alignment method of a normalized time axis to obtain state characteristic values of the intermittent operation equipment in each time interval of the operation stage;

[0011] Determining the degree of operation deviation of the intermittently operated equipment based on the state characteristic values within each time interval and a pre-established normal behavior baseline;

[0012] Whether an abnormality occurs in the intermittent operation equipment is determined based on the degree of operation deviation.

[0013] Optionally, in some embodiments of the present application, the detection data includes a first parameter and a second parameter having characteristic correlation, wherein identifying the detection data based on an operation state identification mechanism of multi-signal fusion to determine the operation stage of the intermittent operation device includes:

[0014] determining a change trend of the first parameter and a change trend of the second parameter;

[0015] The operation phase and the stop phase of the intermittent operation equipment are distinguished based on the change trend of the first parameter, the change trend of the second parameter, and the characteristic correlation between the first parameter and the second parameter.

[0016] Optionally, in some embodiments of the present application, when the first parameter is the motor current signal of the intermittently running device and the second parameter is the vibration signal of the intermittently running device, if the motor current signal is less than a preset current threshold, and the vibration signal is less than a preset vibration threshold, and the duration exceeds a first preset time length, and the downward change trend of the motor current signal is synchronized with the downward change trend of the vibration signal, then it is determined that the intermittently running device has entered the stop stage.

[0017] Optionally, in some embodiments of the present application, processing the detection data of the operation phase based on the period alignment of the normalized time axis includes:

[0018] Determine the start time and end time of the operation phase, and determine the cycle length based on the start time and end time;

[0019] Dividing the cycle duration based on preset time intervals to obtain time intervals of the operation phase;

[0020] Feature extraction is performed on the detection data in each time interval to obtain the state feature value of the intermittent operation equipment in each time interval of the operation stage.

[0021] Optionally, in some embodiments of the present application, the state feature value includes multiple statistical features of the same data.

[0022] Optionally, in some embodiments of the present application, the normal behavior baseline is created according to the following steps:

[0023] Obtain historical data of the intermittently operated equipment during the period;

[0024] Processing the historical data based on the preset time intervals to obtain statistical features within each time interval;

[0025] The normal behavior baseline is created according to the statistical features in each time interval.

[0026] Optionally, in some embodiments of the present application, determining the degree of operation deviation of the intermittently operated device according to the state characteristic values within each time interval and a pre-established normal behavior baseline includes:

[0027] Calculating the deviation between the state feature value in each time interval and the normal behavior baseline;

[0028] Sum the deviation values corresponding to each time interval to obtain the total deviation value;

[0029] The total deviation value is scored for deviation anomaly using a linear function to obtain a deviation anomaly score value.

[0030] In a second aspect, an embodiment of the present application provides an abnormality detection device for intermittently operated equipment, comprising:

[0031] an acquisition module, configured to acquire detection data of the intermittently operated device, wherein the detection data is used to characterize the operating state of the intermittently operated device;

[0032] an identification module, configured to identify the detection data based on an operation status identification mechanism of multi-signal fusion to determine an operation stage of the intermittent operation equipment;

[0033] a processing module, configured to process the detection data of the operation phase based on a period alignment method of a normalized time axis, obtain state characteristic values of the intermittent operation device in each time interval of the operation phase, and determine the degree of operation deviation of the intermittent operation device based on the state characteristic values in each time interval and a pre-established normal behavior baseline;

[0034] The detection module is used to determine whether an abnormality occurs in the intermittent operation equipment according to the degree of operation deviation.

[0035] In a third aspect, an embodiment of the present application provides a computer device, comprising a processor and a memory, wherein the processor is configured to execute instructions stored in the memory so that the computer device performs the above-mentioned method for detecting anomalies in intermittently operated equipment.

[0036] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, comprising computer program instructions, which, when executed by a processor, implement the above-mentioned method for detecting anomalies in intermittently operated equipment.

[0037] (3) Beneficial effects

[0038] The anomaly detection method and device for intermittently running equipment provided in the embodiments of the present application, by introducing an operating status recognition mechanism in the data acquisition stage, can realize automatic recognition and isolation of missing or "flat line" signals generated in the stop stage, prevent misjudgment, and convert non-continuous detection data into an input format suitable for anomaly detection through adaptive segmented feature extraction and mapping. In this way, even in the case of frequent equipment shutdown and sampling interruptions, feature calculation and matching can still be completed independently in each operating segment, ensuring the stability, reliability and accuracy of the overall detection process. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 1 is a flow chart of a method for detecting anomalies in intermittently operated equipment according to one embodiment of the present application;

[0040] Figure 2 A flowchart for identifying an operating phase of an intermittently operated device according to one embodiment of the present application;

[0041] Figure 3 This is a flowchart of processing detection data of the operation phase based on the period alignment method of the normalized time axis according to one embodiment of the present application;

[0042] Figure 4 A flowchart of creating a normal behavior baseline according to one embodiment of the present application;

[0043] Figure 5A flowchart of determining the degree of operation deviation of an intermittently operated device according to one embodiment of the present application;

[0044] Figure 6 Schematic diagram of a block diagram of an abnormality detection device for intermittently operated equipment according to one embodiment of the present application;

[0045] Figure 7 Schematic diagram of abnormality detection for intermittently operated equipment according to one embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to better explain the present application and facilitate understanding, the present application is described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0047] In related technologies, anomaly detection for intermittently running equipment can be mainly classified into three categories:

[0048] The first category is anomaly detection solutions based on the statistical analysis of continuous time series. This solution extracts statistical features such as mean, variance, standard deviation, trend, or spectrum from key physical quantities such as current, vibration, and temperature collected during the equipment's operation phase, and compares the results with a pre-established normal state model or preset threshold to determine whether the equipment is abnormal. However, this solution is highly dependent on the continuity of data. If zero current or a "flat line" signal appears during the equipment's shutdown phase, the algorithm loses its analytical basis or misjudges a normal shutdown as a fault. At the same time, for intermittently operating equipment with short operating cycles and scarce samples, the early fault characteristics are difficult to capture in a timely manner, resulting in a high rate of missed reports.

[0049] The second category is a sequence reconstruction and anomaly detection solution based on deep learning. This solution uses neural network models such as autoencoders and LSTM to perform encoding and decoding training on time series data of normal equipment operation, and uses the error between the input data and the model reconstruction results to determine whether the equipment has an anomaly. Although this method can theoretically handle non-stationary and intermittent sequences, it is highly dependent on a large number of high-quality training samples. The scarcity of data caused by the short cycle of intermittent operation of equipment makes model training insufficient. In addition, the complex network structure and high computing power requirements make it difficult to deploy to on-site edge devices. In addition, there is a lack of special preprocessing for completely "breakpoint" data. The reconstruction error is often interfered with by the shutdown signal, resulting in false alarms.

[0050] The third category is anomaly detection solutions based on interpolation / missing value processing. This solution fills in missing data segments during periods of equipment downtime with linear interpolation, spline interpolation, or Gaussian process prediction, restoring the data to a "continuous" sequence. Traditional statistical analysis or classification models are then applied to determine anomalies. While this approach can temporarily satisfy the continuity assumption, the "virtual data" generated by interpolation or prediction deviates significantly from the actual operating signal, and these errors are often transmitted to subsequent detection steps, leading to false positives or missed detections. Furthermore, the selection and parameter tuning of different interpolation models are difficult and lack universal applicability. The masking of breakpoint features also weakens sensitivity to intermittent characteristics.

[0051] To this end, the abnormality detection method and device for intermittently running equipment provided in the embodiments of the present application, by introducing an operating status recognition mechanism in the data acquisition stage, can realize automatic recognition and isolation of missing or "flat line" signals generated in the stop stage, prevent misjudgment, and convert non-continuous detection data into an input format suitable for abnormality detection through adaptive segmented feature extraction and mapping. In this way, even in the case of frequent equipment shutdown and sampling interruption, feature calculation and matching can still be completed independently in each operating segment, ensuring the stability, reliability and accuracy of the overall detection process. Not only is the detection method simple, but the detection accuracy is high and the versatility is strong.

[0052] To better understand the above technical solutions, exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a clearer and more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0053] Figure 1 FIG. 1 is a flow chart of an abnormality detection method for intermittently operated equipment according to an embodiment of the present application. Figure 1 As shown, the abnormality detection method of the intermittent operation equipment includes:

[0054] S1, acquiring detection data of an intermittently operated device, wherein the detection data is used to characterize an operating state of the intermittently operated device.

[0055] Intermittent equipment, as it's understood, refers to equipment in fields like industrial production and energy extraction that doesn't operate continuously but instead periodically starts and stops based on production needs. Examples include pumping units and water injection pumps in oil and gas field production. This operating mode typically results in discontinuous data generated by the equipment.

[0056] In some embodiments of the present application, various types of sensors can be placed at key locations on intermittently operating equipment. For example, current sensors can be installed on motor windings to monitor the motor's operating current in real time; vibration sensors can be installed on vibration-sensitive locations on intermittently operating equipment to obtain vibration information during operation; and voltage sensors can be installed on the power supply circuit of intermittently operating equipment to monitor voltage information during operation. These sensors can accurately capture various physical signals during equipment operation.

[0057] Based on these sensors, key data such as the current signal I(t), vibration signal A(t), and voltage signal V(t) during the operation of the equipment can be continuously detected. However, in the actual data acquisition process, the raw data detected by these sensors are often interfered with by various noises, such as electromagnetic noise, environmental noise, etc. These noises will affect the accuracy and availability of the data, thereby interfering with subsequent abnormality detection and analysis. Therefore, in some embodiments of the present application, the data detected by the sensor needs to be preprocessed. For example, in order to remove noise interference and improve data quality, a digital filtering algorithm can be used to preprocess the raw data. Among them, taking the low-pass filter as an example, it can effectively suppress high-frequency noise and retain the low-frequency useful components of the signal. Assuming that the original signal is x(t), the signal y(t) after the low-pass filter can be expressed by the following convolution formula:

[0058]

[0059] Here, h(n) is the unit impulse response of the low-pass filter, which determines the filter's attenuation characteristics for signals of different frequencies. N is the filter order. A higher order yields a finer filtering effect, but also increases computational complexity. In practical applications, the values of h(n) and N must be appropriately selected based on the characteristics of the device's signal and noise profile to achieve optimal filtering results.

[0060] Specifically, in one example of this application, a Butterworth low-pass filter can be used to pre-process the original collected signal, and the order and cut-off frequency of the filter can be reasonably set to remove high-frequency noise and transient interference, significantly improving data quality and availability, and providing a basis for device anomaly detection.

[0061] S2 uses a multi-signal fusion operating status recognition mechanism to identify the operating phase of intermittently operating equipment using detection data. This data is then processed based on the periodic alignment of the normalized time axis to obtain the state characteristic values of the intermittently operating equipment at each time interval during the operating phase, providing a stable data foundation.

[0062] S21, determining a change trend of the first parameter and a change trend of the second parameter.

[0063] S22 , distinguishing between the operation phase and the stop phase of the intermittent operation equipment based on the change trend of the first parameter, the change trend of the second parameter, and the characteristic correlation between the first parameter and the second parameter.

[0064] It should be noted that the first parameter and the second parameter are two related parameters, such as current parameter and vibration parameter, which are coupled with each other when the device is running. They are usually two key parameters that are positively correlated when characterizing the operating status of the device.

[0065] Specifically, when the first parameter is the motor current signal of the intermittently running equipment and the second parameter is the vibration signal of the intermittently running equipment, if the motor current signal is less than the preset current threshold, and the vibration signal is less than the preset vibration threshold, and the duration exceeds the first preset time length, and the downward change trend of the motor current signal is synchronized with the downward change trend of the vibration signal, it is determined that the intermittently running equipment has entered the stop stage.

[0066] That is to say, in the embodiment of the present application, when constructing a multi-signal fusion operation status identification mechanism, the feature correlations of multiple key feature signals and the multi-dimensional time-frequency features are fused, and the relevant thresholds can also be dynamically adjusted, so as to accurately identify whether the equipment has entered a stop period and avoid interference with abnormal deviations caused by faults in the equipment.

[0067] Among them, the multi-signal fusion operation status recognition mechanism can establish reference templates in the equipment operation stage and the stop stage respectively, and use the transition characteristics of the equipment stop and start transient signals (such as the current rise rate, the change of vibration harmonic components, etc.) for comparison, and combined with the statistical distribution of historical normal shutdown data, it can effectively shield the false alarms caused by routine shutdowns, so that anomaly detection only issues alarms for real fault signals, reducing the cost of on-site operation and maintenance intervention.

[0068] Optionally, in a specific example of the present application, the pre-processed current signal I(t) and vibration signal A(t) are selected as key characteristic signals for judging the operating status of the device. This is because the current and vibration conditions can intuitively and effectively reflect the operating status of the device. When the device is running, the current and vibration are usually within a certain range and fluctuate regularly, while when it stops, there will be obvious changes. In addition, in order to accurately distinguish the operating stage and the stopping stage of the device, a preset current threshold I can be set. th , preset vibration threshold A th And the first preset time T stop These thresholds and time parameters can be determined through multiple tests and optimizations based on the analysis of a large amount of historical operating data of the equipment, combined with the working principle and actual operating conditions of the equipment. If the current signal I(t) is always less than I th, and the vibration signal A(t) is less than A th , and the state lasts for time t duration Exceeding the preset T stop , and the decreasing trend of the current signal I(t) keeps synchronization with the decreasing trend of the vibration signal A(t), it can be determined that the equipment is in the stopping stage; otherwise, as long as I(t)≥I th And A(t)≥A th , it is determined that the equipment is in the operation stage.

[0069] This precise judgment rule can effectively avoid misjudging the normal stop state of the equipment as an abnormality, greatly improving the accuracy of abnormality detection.

[0070] The cycle alignment method based on the normalized time axis means that the detection data of each operating cycle of the intermittently running equipment is divided into an equal number of time intervals according to relative time. In this way, regardless of the actual duration of each cycle, features are extracted at these uniformly divided time intervals, thereby achieving alignment of data from different operating cycles on the same time scale, which facilitates cross-cycle feature comparison and analysis, ensuring improved anomaly detection accuracy.

[0071] Optionally, in some embodiments of the present application, Figure 3 As shown, the detection data of the operation phase is processed based on the period alignment method of the normalized time axis, including:

[0072] A ring buffer is used to store the periodic data of the current cycle and is updated in real time according to the sampling points. When a cycle end mark is detected (such as a speed signal trigger or time arrival), the cycle duration is immediately divided based on a preset time interval to obtain the various time intervals of the operating stage; feature extraction is performed on the detection data within the various time intervals to obtain the state feature values of the intermittently running equipment within the various time intervals of the operating stage.

[0073] State characteristic values include: time domain characteristics (mean, variance, kurtosis, waveform factor), frequency domain characteristics and time-frequency domain characteristics.

[0074] Taking the current signal as an example, the waveform factor is calculated using the following expression: Where RMS is the root mean square value;

[0075] Frequency-domain features are generated by zero-padding the periodic data to a power-of-two length (e.g., 256 points) and then computing a fast Fourier transform to obtain a spectrum. The amplitudes of the first M harmonics (e.g., the fundamental frequency, the second harmonic, the third harmonic, etc.) are extracted from this spectrum, along with the corresponding harmonic distortion (THD).

[0076] Finally, the time-frequency domain features are characterized by the centroid of the time-frequency domain energy spectrum. This is obtained by the following steps:

[0077] STFT analysis: Divide the periodic data into frames (e.g., 64 points per frame, 50% overlap), and add a Hanning window.

[0078] S31, determining the start time and end time of the operation phase, and determining the cycle duration according to the start time and the end time.

[0079] S32: Divide the cycle duration based on preset time intervals to obtain time intervals of the operation phase.

[0080] S33, performing feature extraction on the detection data in each time interval to obtain the state feature value of the intermittent operation equipment in each time interval during the operation phase.

[0081] Specifically, after the operation status recognition mechanism based on multi-signal fusion determines that the equipment is in the operation stage, the starting time t of each operation cycle is determined. start and end time t end , thus accurately calculating the cycle length T = t end -t start Since the operating cycle of intermittent equipment may vary due to factors such as production demand and equipment status, it is difficult to compare and analyze data from different cycles. In order to solve the cycle difference problem, this application can divide each operating cycle into P time intervals, ensuring that the length of each time interval is Uniformity is achieved, so that no matter how the actual duration of different operating cycles changes, they can be unified into the same time scale for analysis. For each time interval i (i=1,2,…,P), in the corresponding time interval ([t start +(i-1)Δt,t start +iΔt]) to extract the characteristic value of the signal. Taking the current signal as an example, calculate the mean value μ in this time period i , can be accurately calculated by the following integral formula:

[0082]

[0083] This mean μ i This effectively represents the characteristics of the current signal within this time interval. This application uses this normalized time axis cycle alignment method to align data from different operating cycles in the time dimension. This allows cross-cycle feature comparison and analysis to more accurately detect potential anomalies during device operation.

[0084] It can be seen from this that the anomaly detection method for intermittently running equipment in the embodiment of the present application can eliminate the strict dependence on continuous sequences, and realize automatic identification and isolation of missing or "flat line" signals generated during the downtime by introducing the "run / stop" breakpoint identification and segmentation preprocessing mechanism in the data acquisition layer. The downtime segment data is neither simply discarded nor involved in the continuity analysis. Instead, adaptive segmented feature extraction and mapping are realized through the periodic alignment method based on the normalized time axis, thereby converting the non-continuous monitoring data into an input format suitable for anomaly detection. In this way, even in the scenario where the equipment is frequently stopped and started and the sampling is interrupted, feature calculation and matching can still be completed independently in each operating segment to ensure that the overall detection process is robust and reliable. Through the coordination of the operating status recognition mechanism of multi-signal fusion and the periodic alignment method based on the normalized time axis, the data processing process is simplified and the accuracy of anomaly detection is ensured.

[0085] Optionally, in some embodiments of the present application, the state characteristic value may include multiple statistical features of the same data, such as mean, variance, spectrum, etc. Through the comprehensive judgment of multiple statistical features, the signal characteristics of each time interval under normal operating conditions of the device can be fully reflected, thereby effectively improving the accuracy of anomaly detection.

[0086] Optionally, after extracting features from the detection data within each time interval, the method further includes:

[0087] Based on the wavelet packet energy entropy-operating condition adaptation (WPEE-CA) fusion algorithm, the extracted time domain and frequency domain features are dynamically processed to generate a comprehensive feature vector. Specifically including:

[0088] Perform wavelet packet decomposition on each periodic data to obtain sub-signals of multiple sub-frequency bands;

[0089] Calculate the energy entropy of each frequency band, the expression is:

[0090]

[0091] Among them, p l,k is the energy proportion of the lth frequency band in the kth time window, K is the number of time windows;

[0092] The weight w is assigned by the adaptive weight formula through the acquired detection data C (current signal, vibration signal, etc.) l ; Its expression is:

[0093]

[0094] Among them, γ is the adjustment parameter, L is the total number of frequency bands;

[0095] Generate comprehensive feature vector F i :

[0096]

[0097] Among them, f m,l is the characteristic value of the lth frequency band in the mth time interval.

[0098] Specifically, wavelet packet decomposition is performed on each periodic data to obtain sub-signals of multiple sub-bands, including:

[0099] Normalize the detection data (such as current and vibration) of each operating cycle to eliminate dimensionality effects. Use a low-pass filter (such as a Butterworth filter) to remove high-frequency noise and retain the effective frequency band signal.

[0100] Determine the basis function and frequency band resolution of the decomposition: Select a suitable mother wavelet (such as Daubechies, Symlet) according to the signal characteristics. DB4 or SYM6 are commonly used because they have a good balance between time-frequency localization and computational efficiency. Select the number of layers R according to the target frequency band resolution. The number of frequency bands after decomposition is 2. R .

[0101] After the Rth layer decomposition, the lth frequency band (l=1,2,……2 R ) frequency band range is:

[0102] Among them, f s is the signal sampling rate.

[0103] Wavelet packet decomposition is performed to decompose the signal layer by layer into sub-signals of different frequency bands. The decomposition algorithm is to initialize the detection data as the root node (layer 0).

[0104] Each node (frequency band) is low-pass and high-pass filtered to generate two child nodes (low-frequency approximation number A and high-frequency detail coefficient D), and the process is repeated until the target number of layers R is reached.

[0105] The sub-signals of each frequency band are reconstructed from the decomposition coefficients, and the coefficients of each leaf node (R-th layer node) are inversely wavelet transformed to obtain the sub-signals of the corresponding frequency band. All sub-signals cover the frequency band range of the entire original signal (detection data) and do not overlap with each other.

[0106] In addition, when extracting the state characteristic values within each time interval, this application can also improve the ability to capture short-cycle fault features through a lightweight algorithm that combines "single-cycle fast feature extraction" with "sliding fragment aggregation" to achieve early warning. Among them, in order to address the problem that each operating cycle is limited and fault signs are easily overwhelmed by short-term data, by starting multi-scale feature extraction (time domain statistics, frequency domain harmonic components, time-frequency domain energy spectrum center of gravity, etc.) at the beginning of the equipment operation phase, and performing online aggregation and comparison with the local features of multiple adjacent cycles, potential abnormal offsets can be captured in a short time, so that there is no need to rely on long sequence reconstruction or a large amount of historical data. An early warning can be issued at the early stage of the fault, providing valuable response time for maintenance decisions.

[0107] Furthermore,

[0108] S3, determining the degree of operation deviation of the intermittently operated equipment according to the state characteristic values in each time interval and the pre-created normal behavior baseline.

[0109] Among them, the normal behavior baseline refers to the use of data from multiple historical normal operation cycles of the equipment to calculate the statistical characteristics (such as mean, variance, spectrum, etc.) within each time interval to build a model to represent the normal operating status of the equipment. As the operating conditions of the equipment change, the historical data can be continuously updated and these statistical characteristics can be recalculated, so that the model can dynamically reflect the current normal status of the equipment.

[0110] In some embodiments of the present application, Figure 4 As shown, a normal behavior baseline can be created by following these steps:

[0111] Furthermore, the normal behavior baseline is obtained by using a sliding window to reduce computational complexity and preserve key frequency bands. Specifically, N = 5 historical cycles are selected (dynamically adjusted based on noise levels). A queue (FIFO) is used to store feature vectors within the window, and the oldest data is eliminated when a new cycle arrives. This approach reduces data storage pressure while ensuring real-time updates of the normal behavior baseline and the accuracy of subsequent detection results.

[0112] S41, obtaining historical data of the cycle length of the intermittent operation equipment.

[0113] S42: Process the historical data based on preset time intervals to obtain statistical features within each time interval.

[0114] S43: Create a normal behavior baseline based on the statistical features in each time interval.

[0115] Specifically, a large amount of historical normal operation cycle data of the equipment is obtained. These data are like a "sample library" of the normal operation of the equipment, recording various operating characteristics of the equipment under normal working conditions. Then, after processing the historical data in the above-mentioned cycle alignment method, for each time interval i, various statistical characteristics such as the mean are calculated in detail. variance And spectrum characteristics, etc. Taking mean value calculation as an example, it can be accurately obtained through the following formula:

[0116]

[0117] Where M is the number of historical normal operation cycles, μ i,j is the mean of the i-th time interval in the j-th historical normal operation cycle. The calculation of reflects the discrete degree of the data within the time interval, and the formula can be:

[0118]

[0119] These statistical features comprehensively characterize the signal characteristics of the device at each time interval during normal operation, and together they constitute the normal behavior baseline. Furthermore, given that device operating conditions may change over time due to factors such as equipment aging and production process adjustments, to ensure that the normal behavior baseline accurately reflects the current normal operation of the device, historical normal operation data is regularly updated and these statistical features are recalculated as the device operates. For example, historical data can be updated and statistical features recalculated every 100 cycles, updating the normal behavior baseline and ensuring its real-time validity and accuracy.

[0120] Optionally, in some embodiments of the present application, Figure 5 As shown, the degree of deviation of the intermittent operation equipment is determined based on the status characteristic values in each time interval and the pre-established normal behavior baseline, including:

[0121] S51, calculating the deviation between the state feature value and the normal behavior baseline in each time interval.

[0122] S52: Sum the deviation values corresponding to the respective time intervals to obtain a total deviation value.

[0123] S53, using a linear function to score the total deviation value for deviation anomaly, to obtain a deviation anomaly score value.

[0124] In other words, this application can implement anomaly detection using a cycle-level anomaly matching and scoring mechanism. For the device's current operating cycle, the deviation from the normal behavior baseline is calculated at each time interval. The deviations for all time intervals are accumulated and summed, and then converted into an anomaly score using a predefined scoring function. This score can then be used to determine whether the device is experiencing an anomaly and assess its severity.

[0125] S4, judging whether an abnormality occurs in the intermittent operation equipment according to the degree of operation deviation.

[0126] Specifically, for the current operation cycle of the intermittent operation equipment, data processing is performed strictly according to the cycle alignment method of the normalized time axis, and the corresponding characteristic values are extracted, such as the current mean μ at each time interval. i and variance And, for each time interval i, the Euclidean distance d can be used i To accurately measure the degree of deviation between the characteristic value of the interval and the characteristic value of the interval corresponding to the normal behavior baseline. Taking the mean and variance of the current signal as an example, the Euclidean distance calculation formula is:

[0127]

[0128] This formula comprehensively considers the differences in mean and variance, and can fully reflect the characteristic deviation of the current operation cycle from the normal behavior baseline at each time interval.

[0129] Then the deviation degree d of all time intervals i Accumulate and sum to get the total deviation value D:

[0130]

[0131] Finally, the total deviation value D is converted into an abnormality score S by setting the scoring function S = f (D). For example, a linear function can be used Among them D max is the maximum deviation value calculated from historical data. In this way, no matter how large the absolute value of D is, it can be mapped to the relative range of (0-1), which is convenient for intuitive comparison and judgment.

[0132] In this way, by setting a reasonable alarm threshold S th This threshold can also be determined by analyzing a large amount of historical data and actual operating experience. th When an abnormality occurs, an alarm is triggered immediately, clearly indicating that the equipment may have an abnormality so that the staff can take appropriate measures in time.

[0133] To sum up, the anomaly detection method for intermittently running equipment in the embodiment of the present application combines running / stopping state recognition, cycle normalization alignment, normal behavior baseline modeling, and cycle-level anomaly matching and scoring mechanism to realize anomaly alarm, greatly improve the accuracy and reliability of equipment anomaly detection, and ensure stable production operation.

[0134] Figure 6 FIG. 1 is a block diagram of an abnormality detection device for intermittently operated equipment according to an embodiment of the present application. Figure 6 As shown, the anomaly detection device 600 for intermittently operated equipment includes: an acquisition module 601, an identification module 602, a processing module 603, and a detection module 604. The acquisition module 601 is used to acquire detection data from the intermittently operated equipment, wherein the detection data is used to characterize the operating status of the intermittently operated equipment; the identification module 602 is used to identify the detection data based on an operating status identification mechanism using multi-signal fusion to determine the operating stage of the intermittently operated equipment; the processing module 603 is used to process the detection data from the operating stage based on a periodic alignment method of a normalized time axis, obtain state feature values of the intermittently operated equipment in each time interval of the operating stage, and determine the degree of operational deviation of the intermittently operated equipment based on the state feature values in each time interval and a pre-established normal behavior baseline; and the detection module 604 is used to determine whether an anomaly has occurred in the intermittently operated equipment based on the degree of operational deviation.

[0135] Optionally, in some embodiments of the present application, the detection data includes a first parameter and a second parameter having characteristic correlation, and the identification module 602 is also used to determine the changing trend of the first parameter and the changing trend of the second parameter, and distinguish the running stage and the stopping stage of the intermittently running equipment based on the changing trend of the first parameter, the changing trend of the second parameter, and the characteristic correlation between the first parameter and the second parameter.

[0136] Among them, when the first parameter is the motor current signal of the intermittently running equipment and the second parameter is the vibration signal of the intermittently running equipment, if the motor current signal is less than the preset current threshold, and the vibration signal is less than the preset vibration threshold, and the duration exceeds the first preset time length, and the downward change trend of the motor current signal is synchronized with the downward change trend of the vibration signal, it is determined that the intermittently running equipment has entered the stop stage.

[0137] Optionally, in some embodiments of the present application, the processing module 603 is also used to determine the start time and end time of the operation phase, and determine the cycle duration based on the start time and end time, and divide the cycle duration based on a preset time interval to obtain each time interval of the operation phase, and then perform feature extraction on the detection data within each time interval to obtain the state characteristic value of the intermittent operation equipment within each time interval of the operation phase.

[0138] In addition, the processing module 603 is also used to calculate the deviation value between the state characteristic value and the normal behavior baseline in each time interval, and sum the deviation values corresponding to each time interval to obtain a total deviation value, and use a linear function to score the deviation abnormality of the total deviation value to obtain a deviation abnormality score value.

[0139] The further functional description of the above modules and units is the same as that of the corresponding embodiment of the above method, and will not be repeated here.

[0140] Specifically, in some embodiments of the present application, the abnormality detection process of the abnormality detection device of the intermittent operation equipment can be as follows: Figure 7 As shown, in view of the characteristics of intermittent data discontinuity and obvious periodicity of operation mode of intermittent operation equipment, a data acquisition strategy based on multi-source fusion is used to select different types of sensors such as current, vibration, and pressure, and optimize their installation positions according to the equipment structure and signal characteristics to achieve high-quality data acquisition of the entire equipment operation process. In order to improve the stability of the signal and the effectiveness of subsequent analysis, digital filtering algorithms such as Butterworth low-pass filter are used to pre-process the original collected signal, and the order and cutoff frequency of the filter are reasonably set to remove high-frequency noise and transient interference, which significantly improves the data quality and availability, ensures the accuracy of subsequent data analysis and the improvement of anomaly detection accuracy. In addition, in order to unify the data structure of different operation cycles and improve the temporal alignment and expression ability of features, this application adopts a period alignment method based on the normalized time axis, divides each operation cycle into a fixed number of time intervals, and extracts representative statistical features in each time interval, such as current mean, vibration energy, pressure change rate, etc., to capture the state changes of the equipment in different operation stages. This time alignment and segmented extraction strategy can effectively eliminate the impact of cycle length differences, making the extracted features more comparable and generalizable, and laying a solid foundation for state recognition and anomaly detection. For the run-stop mode of intermittently running equipment, an automated state recognition mechanism based on multi-signal fusion is adopted. Based on the change characteristics of key signals such as current and vibration, the current threshold I is set. th , vibration threshold A th And the first preset time T stop, combined with duration and multi-dimensional signal judgment logic, accurately distinguish between running and stopped states. This mechanism effectively prevents misjudgments caused by instantaneous disturbances or periodic fluctuations, and can flexibly adapt to different types of equipment and working conditions. It is a key prerequisite for achieving cycle extraction and subsequent modeling. At the same time, large-scale historical normal cycle data is used to construct a normal behavior baseline for the normal operation of the equipment. By calculating the statistical features such as the mean and variance of each time interval on the normalized time axis, a high-resolution state template is formed. During the operation of the equipment, the system continuously monitors changes in operating data and sets a dynamic update mechanism based on indicators such as the number of operating cycles and statistical deviations. The normal behavior baseline is periodically adjusted to cope with long-term trend changes such as equipment wear and output changes, ensuring the timeliness and adaptability of the normal behavior baseline, thereby significantly improving the accuracy and robustness of anomaly identification. Finally, on the basis of realizing the normal behavior baseline, this application is based on an anomaly assessment mechanism based on multi-dimensional feature deviation. By calculating the Euclidean distance between the current operating cycle feature vector and the normal behavior baseline, the overall deviation degree is measured, and the deviation of each dimension is summed to generate an anomaly score. The system sets alarm thresholds based on historical data and operational requirements. When the anomaly score exceeds this threshold, an alarm is triggered, alerting maintenance personnel to potential anomalies. This evaluation mechanism, integrating multi-dimensional status information, effectively reduces false positives and false negatives, improving the effectiveness and accuracy of predictive maintenance.

[0141] In addition, an embodiment of the present application further proposes a computer device, which includes a processor and a memory, wherein the processor is used to execute instructions stored in the memory so that the computer device executes the abnormality detection method for intermittently running equipment described in the above embodiment.

[0142] Finally, an embodiment of the present application also proposes a computer-readable storage medium, including computer program instructions, which, when executed by a processor, implements the abnormality detection method for intermittently running equipment described in the above embodiment.

[0143] In summary, the anomaly detection method and device for intermittently running equipment in the embodiment of the present application, based on the automatic identification mechanism of the operating status of multiple signals fusion, can accurately determine the current state of the equipment, effectively avoid false alarms caused by missing data, and unify each operating cycle into an equally spaced time structure through the period alignment method of the normalized time axis, ensuring the consistency and comparability of feature extraction, fundamentally overcoming the negative impact of data discontinuity on model training and anomaly identification, and greatly improving the robustness and stability of the system. In view of the problem that intermittently running equipment has a short operating cycle and discontinuous signals, it is difficult to accumulate enough features to identify initial faults in a short time. Through normalized cycle division and multi-dimensional feature extraction, a complete operating cycle feature representation is established, and then through normal behavior baseline modeling and cycle-level anomaly matching scoring, subtle deviations between current operating data and normal behavior are accurately identified. Even if the equipment only has slight fluctuations, early signs of faults can be identified in time, thereby triggering maintenance responses in advance, effectively reducing production stoppage losses caused by sudden failures. At the same time, by introducing a dynamic update mechanism for the normal behavior baseline and periodically reconstructing statistical features based on real-time collected normal operation data, the normal behavior baseline can be continuously adjusted as equipment operating conditions change. The cycle-level anomaly scoring mechanism can calculate the difference based on the current latest baseline, ensuring that detection accuracy does not decrease due to changes in operating conditions. Therefore, this application can greatly improve the accuracy and reliability of equipment anomaly detection and ensure stable production operations.

[0144] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0145] In this application, unless otherwise specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they can refer to fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in this application based on the specific circumstances.

[0146] In this application, unless otherwise expressly specified or limited, when a first feature is “on” or “below” a second feature, it may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. Moreover, when a first feature is “above”, “above”, or “above” a second feature, it may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is at a higher level than the second feature. When a first feature is “below”, “below”, or “below” a second feature, it may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is at a lower level than the second feature.

[0147] In the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples without contradiction.

[0148] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for detecting abnormalities in intermittently operated equipment, characterized in that: include: Acquiring detection data of the intermittently operated equipment, wherein the detection data is used to characterize the operating status of the intermittently operated equipment; Identifying the detection data based on a multi-signal fusion operation state identification mechanism to determine the operation stage of the intermittent operation equipment, and processing the detection data of the operation stage based on a period alignment method of a normalized time axis to obtain state characteristic values of the intermittent operation equipment in each time interval of the operation stage; Determining the degree of operation deviation of the intermittently operated equipment based on the state characteristic values within each time interval and a pre-established normal behavior baseline; Whether an abnormality occurs in the intermittent operation equipment is determined based on the degree of operation deviation.

2. The method for detecting abnormality of intermittently operated equipment according to claim 1, characterized in that: The detection data includes a first parameter and a second parameter having characteristic correlation, wherein the detection data is identified based on an operation state identification mechanism of multi-signal fusion to determine the operation stage of the intermittent operation equipment, including: determining a change trend of the first parameter and a change trend of the second parameter; The operation phase and the stop phase of the intermittent operation equipment are distinguished based on the change trend of the first parameter, the change trend of the second parameter, and the characteristic correlation between the first parameter and the second parameter.

3. The method for detecting abnormality of intermittently operated equipment according to claim 2, characterized in that: When the first parameter is the motor current signal of the intermittently running device and the second parameter is the vibration signal of the intermittently running device, if the motor current signal is less than the preset current threshold, and the vibration signal is less than the preset vibration threshold, and the duration exceeds the first preset time length, and the downward change trend of the motor current signal is synchronized with the downward change trend of the vibration signal, it is determined that the intermittently running device enters the stop stage.

4. The method for detecting abnormality of intermittently operated equipment according to any one of claims 1 to 3, characterized in that: The detection data of the operation phase is processed based on the period alignment of the normalized time axis, including: Determine the start time and end time of the operation phase, and determine the cycle length based on the start time and end time; Dividing the cycle duration based on preset time intervals to obtain time intervals of the operation phase; Feature extraction is performed on the detection data in each time interval to obtain the state feature value of the intermittent operation equipment in each time interval of the operation stage.

5. The method for detecting abnormality of intermittently operated equipment according to claim 4, characterized in that: The state feature value includes multiple statistical features of the same type of data.

6. The method for detecting abnormality of intermittently operated equipment according to claim 4, characterized in that: The normal behavior baseline is created according to the following steps: Obtain historical data of the intermittently operated equipment during the period; Processing the historical data based on the preset time intervals to obtain statistical features within each time interval; The normal behavior baseline is created according to the statistical features in each time interval.

7. The method for detecting abnormality of intermittently operated equipment according to claim 4, characterized in that: Determining the degree of operation deviation of the intermittently operated equipment according to the state characteristic values in each time interval and a pre-established normal behavior baseline includes: Calculating the deviation between the state feature value in each time interval and the normal behavior baseline; Sum the deviation values corresponding to each time interval to obtain the total deviation value; The total deviation value is scored for deviation anomaly using a linear function to obtain a deviation anomaly score value.

8. An abnormality detection device for intermittently operated equipment, characterized in that: include: an acquisition module, configured to acquire detection data of the intermittently operated device, wherein the detection data is used to characterize the operating state of the intermittently operated device; an identification module, configured to identify the detection data based on an operation status identification mechanism of multi-signal fusion to determine an operation stage of the intermittent operation equipment; a processing module, configured to process the detection data of the operation phase based on a period alignment method of a normalized time axis, obtain state characteristic values of the intermittent operation device in each time interval of the operation phase, and determine the degree of operation deviation of the intermittent operation device based on the state characteristic values in each time interval and a pre-established normal behavior baseline; The detection module is used to determine whether an abnormality occurs in the intermittent operation equipment according to the degree of operation deviation.

9. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the processor is configured to execute instructions stored in the memory, so that the computer device executes the abnormality detection method for intermittently operated equipment according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The method comprises computer program instructions, and when the computer program instructions are executed by a processor, the method for detecting abnormality of an intermittently operated device according to any one of claims 1 to 7 is implemented.

Images