Special equipment state real-time detection system and method

Abstract

The invention relates to the technical field of state monitoring, in particular to a special equipment state real-time detection system and method, and the system comprises a data processing module, an association construction module, an abrupt change extraction module, a state recognition module and a trend prediction module. According to the method, by combining signal continuity analysis and denoising processing, the original data quality is improved, the influence of noise interference on state evaluation is reduced, multi-cycle data feature extraction and clustering analysis are utilized, a dynamic correlation model is constructed, the state classification accuracy under different working conditions is enhanced, and a trend direction and fluctuation form combined analysis mechanism is adopted; the method is advantaged in that abnormal signal period characteristics are accurately captured, early-stage identification efficiency of potential faults is improved, a multi-dimensional correlation characteristic comparison verification method is integrated, a composite state discrimination system is established, single-dimension misjudgment risks are reduced, amplification proportion sequence and persistence analysis technologies are fused, and a probabilistic prediction model is combined. And early warning of the abnormal state time node of the special equipment is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of condition monitoring, and in particular to a real-time detection system and method for the condition of special equipment. Background Art

[0002] The technical field of condition monitoring involves the real-time monitoring and detection of the conditions of various devices or machines during operation. The core content is to collect and analyze the operating parameters of the devices in a timely manner to discover potential faults or abnormalities of the devices, and conduct early warning and maintenance. The condition monitoring technology is widely used in industrial production, energy, transportation and other fields. Especially for special equipment, such as pressure vessels, elevators, cranes, etc., it has important significance. Through sensors, detection instruments and data acquisition systems, it provides a comprehensive and accurate assessment of the operating state of the equipment, ensures the safe operation of the equipment, and reduces the fault downtime and maintenance costs.

[0003] Among them, the real-time detection system for the condition of special equipment refers to a method and system solution for real-time collection and identification of the operating state information of special equipment. The patent aims at the problems that the key state parameters of special equipment are invisible during operation and the change process is difficult to judge, and constructs a systematic technical path composed of multi-point signal acquisition, data processing and state analysis. The system includes collecting multi-dimensional physical quantities by arranging temperature sensors, pressure sensors and vibration sensors, completing the digital processing of signals through an analog-to-digital conversion circuit, using a built-in processing module to classify numerical features, calculate statistical models and judge state classification, and uploading the recognition results to the monitoring platform through wireless communication. It involves multiple links such as sensor acquisition process design, data preprocessing step setting, state classification model definition, communication protocol configuration, and system integration and deployment.

[0004] The traditional real-time detection technology for the condition of special equipment has limited ability to identify abnormal jump signals. The instantaneous interference in the original data is easily misjudged as a real state change, resulting in a decrease in the credibility of subsequent analysis. Feature analysis is mostly based on static statistics of a single time window, lacking the associated modeling of cross-cycle feature combinations, and it is difficult to capture the internal connection between the gradual change process and the mutation signs of the equipment operating state. The state classification model depends on preset thresholds and manual experience rules. When the equipment operating environment changes or components age, the model's self-adaptive adjustment ability is insufficient and prone to classification deviation. Abnormal detection mainly conducts independent analysis for a single signal dimension, without establishing a comprehensive analysis mechanism for the co-variation of multi-physical quantities, missing composite fault features. The communication transmission link focuses on result reporting, without deeply integrating real-time data with the prediction model, resulting in a lag in state warning and causing potential safety hazards. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art, and to propose a real-time detection system and method for the condition of special equipment.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A real-time detection system for the state of special equipment includes:

[0007] The data processing module acquires the sensor monitoring data, including temperature signals, pressure signals, and vibration signals. By analyzing the continuity of data changes, it detects and eliminates abnormal jump signals, denoises the data, and generates a denoised signal feature set.

[0008] The association construction module extracts the data features of each type of signal within multiple time periods according to the denoised signal feature set, induces the co-occurrence patterns of signal feature combinations through cluster analysis, identifies the associated feature groups and extracts the discrimination rules, constructs the feature structure relationship between the operation data, and generates an operation state classification model.

[0009] The mutation extraction module calls the operation state classification model. By analyzing the operation data sequence of the equipment in real time, it extracts the trend direction change and amplitude change in each period, identifies the signal period with abnormal changes, and analyzes the change pattern features of the target signal segment, including trend continuity and fluctuation form, to generate a mutation-driven feature unit.

[0010] The state recognition module calls the operation state classification model and the mutation-driven feature unit, extracts the data features of abnormal signals, and identifies the operation state of the equipment by comparing with the data features of multiple associated feature groups, generating a state feature classification label.

[0011] As a further solution of the present invention, the denoised signal feature set includes signal change trend labels, cycle stability identifiers, and abnormal segment markers. The operation state classification model includes a signal feature grouping structure, a state clustering label group, and a classification discrimination factor set. The mutation-driven feature unit is specifically a trend reversal node, an amplitude abnormal section, and a change structure feature block. The state feature classification label includes a state type number, a dominant signal feature item, and a linked signal classification item.

[0012] As a further solution of the present invention, the data processing module includes:

[0013] The data acquisition sub-module acquires the sensor monitoring data, collects the temperature signals, pressure signals, and vibration signals during the operation of the special equipment, and generates the original equipment operation data.

[0014] The signal extraction sub-module calls the original equipment operation data, extracts the fluctuation amplitude, cycle average change direction, and extreme value difference of each type of signal, and generates a signal feature data set.

[0015] The data denoising sub-module calls the signal feature data set. By analyzing the continuity of data changes, it uses the formula:

[0016]

[0017] Calculate the jump sensitivity coefficient, compare the coefficient with the jump determination threshold, filter out the abnormal jump points of the signal, and obtain the denoised signal feature set;

[0018] Among them, D jump is the jump sensitivity coefficient, N is the total number of signal data points, x i is the value of the i-th signal data point, x i-1 is the value of the (i - 1)-th signal data point, μ is the mean of the signal data set, i is the index of the current signal data point for calculating the difference between adjacent data points, i - 1 is the index of the data point before the current signal data point i, j is the index in the signal data set for calculating the variance of the signal data, and x j is the value of the j-th signal data point.

[0019] As a further solution of the present invention, the association construction module includes:

[0020] The signal feature extraction sub-module extracts the trend level, fluctuation amplitude, and change direction of the temperature, pressure, and vibration signals in multiple time periods according to the denoised signal feature set, and generates a signal feature set;

[0021] The signal clustering analysis sub-module calls the signal feature set, summarizes the signal feature combinations through clustering analysis, extracts the co-occurrence patterns of the signal feature combinations, and uses the formula:

[0022]

[0023] Calculate the clustering effect, obtain the clustering center eigenvalue, and identify the associated feature groups;

[0024] Among them, J Z represents the sum of the squared errors of the signal feature points to their clustering centroids, which is used to evaluate the clustering effect, K is the number of clusters, N k is the number of signal feature points in the k-th cluster, is the value of the i k -th signal feature in the k-th cluster, μ k is the centroid of the k-th cluster, k is the cluster index, and i k represents the i k -th data point in the k-th cluster;

[0025] The model construction sub-module calls the associated feature groups, extracts the discrimination rules, constructs the feature structure relationship between the device operation data, and generates the operation state classification model.

[0026] As a further solution of the present invention, the mutation extraction module includes:

[0027] The trend change extraction sub-module calls the operating state classification model, extracts the trend direction change and amplitude change in each period by analyzing the real-time operating data sequence of the device, and generates a trend change feature set;

[0028] Based on the trend change feature set, the abnormal period identification sub-module uses the formula:

[0029]

[0030] Calculate the abnormal score of each signal period, compare it with the set threshold, identify the abnormal signal period, and obtain the abnormal signal period label;

[0031] Among them, S abnormal is the abnormal degree score, m is the total number of data points in the current period, q is the index of the qth data point, X q is the signal value of the qth data point, including temperature, pressure, vibration, T avg is the arithmetic mean of all data points in the current period, ΔT q is the difference between the qth data point and the previous point, reflecting the instantaneous amplitude change, ΔD q is the trend direction label of the qth data point;

[0032] The mutation mode analysis sub-module calls the abnormal signal period label to analyze the change mode characteristics of the target signal segment, including trend continuity and fluctuation form, and generates a mutation driving feature unit.

[0033] As a further solution of the present invention, the state recognition module includes:

[0034] The feature extraction sub-module calls the operating state classification model and the mutation driving feature unit, extracts the trend level, fluctuation amplitude, and change direction from the abnormal signal period, and generates a signal feature parameter set;

[0035] Based on the signal feature parameters, the similarity calculation sub-module compares and analyzes the data features in multiple associated feature groups according to the trend level, amplitude value, and direction change amount, and uses the formula:

[0036]

[0037] Calculate the similarity value and obtain the feature matching similarity value;

[0038] Among them, S m′ is the feature matching similarity value, M is the total number of features, w p is the weight coefficient of the pth feature, X m′p is the pth feature value of the signal to be recognized, X pis the p-th feature reference value in the known associated feature group, σ m′ is the fluctuation variance value of the p-th feature of the signal to be recognized, σ p is the fluctuation variance value of the p-th feature in the known associated feature group, α is an adjustment factor, p represents the index of the feature, and m′ represents the feature group index of the signal to be recognized;

[0039] The state classification sub-module, based on the feature matching similarity value, corresponds to the state label in real time to identify the operating state of the device and generate a state feature classification label.

[0040] As a further solution of the present invention, the system further includes:

[0041] The trend prediction module extracts the trend direction markers, increase rate sequences, and trend persistences within multiple periods based on the denoised signal feature set and the state feature classification label, analyzes the signal change trend, calculates the probability of abnormal state occurrence, and predicts the time nodes of the state change section to obtain trend-driven warning nodes;

[0042] The trend-driven warning nodes include the starting index of state change, the trend continuation flag bit, and the warning level parameter.

[0043] As a further solution of the present invention, the trend prediction module includes:

[0044] The trend extraction sub-module obtains the denoised signal feature set and the state feature classification label, judges the direction consistency, calculates the increase difference, and counts the trend maintenance time for the signal within each period, marks the trend direction, increase rate, and trend persistence, and obtains the periodic trend feature set;

[0045] The probability analysis sub-module, according to the periodic trend feature set, uses the formula:

[0046]

[0047] calculates the probability of abnormal state occurrence for each period and generates a periodic abnormal probability sequence;

[0048] where, P′ i′ represents the probability of abnormal state occurrence in the i′-th period, TQ i′ represents the trend direction marker in the i′-th period, SZ i′ represents the increase rate in the i′-th period, DB i′ represents the fluctuation degree value in the i′-th period, VZ i′ represents the vibration amplitude in the i′-th period, RK i′ represents the trend duration in the i′-th period, and i′ is the period number;

[0049] The node positioning sub-module calls the periodic anomaly probability sequence, analyzes the changing trend of the anomaly probability in each period, detects the mutation interval of the anomaly probability, predicts the time nodes of the state change section, and obtains the trend-driven warning nodes.

[0050] A method for real-time detection of the state of special equipment, the method for real-time detection of the state of special equipment is executed based on the above-mentioned real-time detection system for the state of special equipment, and includes the following steps:

[0051] S1: Based on the sensor monitoring data, obtain the temperature, pressure and vibration signals during the operation of the special equipment, detect and eliminate the abnormal jump signals by analyzing the continuity of each signal change, denoise the data, and generate a denoised signal feature set;

[0052] S2: Based on the denoised signal feature set, extract the data features of each type of signal in multiple time periods, combine clustering analysis to summarize the signal feature combinations, identify the co-occurrence patterns between the features, classify the relevant features and extract the discrimination rules, and generate an operating state classification model;

[0053] S3: Based on the operating state classification model, by obtaining the operating data sequence of the equipment in real time, analyze the change of the trend direction and the change of the amplitude in each period, extract the abnormal change period, and analyze the change pattern of the signal in the target period, including trend continuity and fluctuation form, and generate a mutation-driven feature unit;

[0054] S4: Based on the operating state classification model and the mutation-driven feature unit, extract the data features in the abnormal signal, and compare and analyze them with the associated feature groups to identify the current operating state of the equipment and generate a state feature classification label;

[0055] S5: Based on the denoised signal feature set and the state feature classification label, extract the trend direction markers, amplitude increase ratio sequences, and trend persistence of each type of signal in multiple periods, calculate the probability of the occurrence of the abnormal state in each period by analyzing the signal change trend, predict the time nodes of the state change, and obtain the trend-driven warning nodes.

[0056] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0057] In the present invention, by combining signal continuity analysis and denoising processing, the quality of the original data is improved, the influence of noise interference on state evaluation is reduced, a dynamic association model is constructed by using multi-cycle data feature extraction and clustering analysis, the accuracy of state classification under different working conditions is enhanced, a joint analysis mechanism of trend direction and fluctuation form is adopted to accurately capture the periodic characteristics of abnormal signals, the early recognition efficiency of potential faults is improved, a multi-dimensional association feature comparison and verification method is integrated to establish a composite state discrimination system, the risk of misjudgment in a single dimension is reduced, the increase ratio sequence and the persistence analysis technology are fused, and combined with a probabilistic prediction model, an early warning of the time node of the abnormal state is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is the system flow chart of the present invention;

[0059] Figure 2 is the flow chart of the data processing module of the present invention;

[0060] Figure 3 is the flow chart of the association construction module of the present invention;

[0061] Figure 4 is the flow chart of the mutation extraction module of the present invention;

[0062] Figure 5 is the flow chart of the state recognition module of the present invention;

[0063] Figure 6 is the flow chart of the trend prediction module of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0064] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0065] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In addition, in the description of the present invention, the meaning of "a plurality" is two or more, unless otherwise specifically defined.

[0066] Please refer to Figure 1 , the present invention provides a technical solution, a special equipment state real-time detection system includes:

[0067] The data processing module obtains the sensor monitoring data, including temperature signals, pressure signals, and vibration signals. By analyzing the continuity of data changes, it detects and eliminates abnormal jump signals, denoises the data, and generates a denoised signal feature set;

[0068] The association construction module extracts the data features of each type of signal within multiple time periods based on the denoised signal feature set. Through cluster analysis, it summarizes the co-occurrence patterns of signal feature combinations, identifies associated feature groups, and extracts discrimination rules, constructs the feature structure relationship between operation data, and generates an operation state classification model;

[0069] The mutation extraction module calls the operation state classification model. By analyzing the operation data sequence of the device in real time, it extracts the trend direction change and amplitude change in each period, identifies the signal period with abnormal changes, and analyzes the change pattern features of the target signal segment, including trend continuity and fluctuation form, and generates mutation-driven feature units;

[0070] The state recognition module calls the operation state classification model and the mutation-driven feature units, extracts the data features of abnormal signals, and by comparing with the data features of multiple associated feature groups, identifies the operation state of the device and generates state feature classification labels;

[0071] The trend prediction module extracts the trend direction markers, amplitude increase ratio sequences, and trend persistence within multiple periods based on the denoised signal feature set and the state feature classification labels, analyzes the signal change trend, calculates the probability of abnormal state occurrence, and predicts the time nodes of the state change section to obtain trend-driven warning nodes.

[0072] The denoised signal feature set includes signal change trend labels, cycle stability identifiers, and abnormal segment markers. The operation state classification model includes signal feature grouping structures, state clustering label groups, and classification discrimination factor sets. The mutation-driven feature units are specifically trend reversal nodes, amplitude abnormal sections, and change structure feature blocks. The state feature classification labels include state type numbers, dominant signal feature items, and associated signal classification items. The trend-driven warning nodes include state change start indexes, trend continuation flag bits, and warning level parameters.

[0073] Please refer to Figure 2 , the data processing module includes:

[0074] The data acquisition sub-module obtains the sensor monitoring data, collects the temperature signals, pressure signals, and vibration signals during the operation of special equipment, and generates the original equipment operation data;

[0075] The execution process of the data acquisition sub-module first obtains various signal data from multiple sensors on the device. These sensors include temperature sensors, pressure sensors, and vibration sensors, which are used to monitor the temperature, pressure, and vibration status of the device in real time. The temperature sensor installed on the device converts the surface temperature of the device into an electrical signal through the thermoelectric effect. The pressure sensor obtains the working pressure inside the device through a capacitive sensor or a strain gauge sensor. The vibration sensor converts mechanical vibration into an electrical signal through a displacement sensor or an accelerometer. After being collected, these raw signals will be transmitted to the data acquisition system, which monitors and records the signals in real time to generate the original operation dataset of the device. Suppose the collected temperature data is: 21.5°C, 22°C, 21.8°C, 21°C, 20.5°C. The pressure signal and the vibration signal are similar, recording the real-time status of the device through sensors and saving them as corresponding time series data. These data will provide basic support for subsequent analysis. For example, the continuity and fluctuations of the temperature signal can help determine the normal working range of the device and its potential risks. Using the formula:

[0076] x i = f(t);

[0077] where x i is the value of the i-th signal data point, and f(t) is the device status signal collected by the signal sensor at time t. Set t1 = 0, t2 = 1, t3 = 2, t4 = 3, t5 = 4, that is, collect the signal data at 5 time points, and the signal data obtained is: 21.5°C, 22°C, 21.8°C, 21°C, 20.5°C.

[0078] The signal extraction sub-module calls the original device operation data to extract the fluctuation amplitude, the change direction of the average period value, and the extreme value difference of each type of signal, generating a signal feature dataset;

[0079] The signal extraction sub-module extracts the characteristics of various signals by calling the original device operation data, such as the fluctuation amplitude, the change direction of the average period value, the extreme value difference, etc. The fluctuation amplitude is calculated by calculating the difference between adjacent data points to reflect the change amplitude of the signal, which can help evaluate whether the device is in an abnormal fluctuation state. The change direction of the average period value is used to identify the working trend of the device, such as the rising or falling trend of the temperature signal, which is very important for monitoring whether the device has an overheating risk. The extreme value difference refers to the difference between the maximum value and the minimum value in the signal, reflecting the working fluctuation range of the device. For example, if the maximum value of the temperature signal is 22°C and the minimum value is 20.5°C, the extreme value difference is: 22°C - 20.5°C = 1.5°C, which indicates that the temperature fluctuation range of the device is relatively large and further analysis and processing are required. Using the formula:

[0080] Δx = |x i - xi-1 |;

[0081] Calculate the fluctuation amplitude, where Δx is the fluctuation amplitude, and x i is the value of the i-th signal data point, and x i-1 is the value of the (i - 1)-th signal data point.

[0082] Set x1 = 21.5 °C and x2 = 22 °C, and calculate the fluctuation amplitude:

[0083] Δx = |22 - 21.5| = 0.5 °C;

[0084] According to the calculation result of the fluctuation amplitude, it can be found that the temperature change is relatively stable and does not constitute an anomaly.

[0085] The data denoising sub-module calls the signal feature data set. By analyzing the continuity of data changes, the formula is adopted:

[0086]

[0087] Calculate the jump sensitivity coefficient. By comparing the coefficient with the jump determination threshold, filter the abnormal jump points of the signal to obtain the denoised signal feature set;

[0088] where D jump is the jump sensitivity coefficient, N is the total number of signal data points, x i is the value of the i-th signal data point, x i-1 is the value of the (i - 1)-th signal data point, μ is the mean of the signal data set, i is the index of the current signal data point for calculating the difference between adjacent data points, i - 1 is the index of the data point before the current signal data point i, j is the index in the signal data set for calculating the variance of the signal data, and x j is the value of the j-th signal data point;

[0089] The data denoising sub-module further performs data denoising processing by calling the previously generated signal feature data set. During the denoising process, first, it is necessary to calculate the jump sensitivity coefficient D jump , which is used to detect whether there are abnormal jump points in the signal. The calculation of the jump sensitivity coefficient quantifies the stability of the signal by analyzing the change amplitude between adjacent data points and combining the variance of the signal data. Specifically, if the value of the jump sensitivity coefficient is large, it means that the signal changes violently and is abnormal data. The formula is as follows:

[0090]

[0091] where D jump is the jump sensitivity coefficient, N is the total number of signal data points, x i is the value of the i-th signal data point, xi-1 is the value of the (i - 1)-th signal data point, μ is the mean of the signal data set, j is the index in the signal data set, used to calculate the variance of the signal data, |x i - x i-1 | is the difference between adjacent data points, is the variance of the signal data set. Assume the signal data is: x1 = 21.5, x2 = 22, x3 = 21.8, x4 = 21, x5 = 20.5, calculate the mean of the signal data set:

[0092]

[0093] Calculate the difference between each pair of adjacent data points:

[0094] |x2 - x1| = |22 - 21.5| = 0.5;

[0095] |x3 - x2| = |21.8 - 22| = 0.2;

[0096] |x4 - x3| = |21 - 21.8| = 0.8;

[0097] |x5 - x4| = |20.5 - 21| = 0.5;

[0098] Calculate the variance of the signal:

[0099]

[0100] Calculate the jump sensitivity coefficient:

[0101]

[0102] According to the calculation result D jump = 4.334, if the set jump determination threshold is 2°C, then this data is an outlier, and these data points should be removed, and finally a denoised signal feature set is obtained.

[0103] Please refer to Figure 3 , the association construction module includes:

[0104] The signal feature extraction sub-module extracts the trend level, fluctuation amplitude, and change direction of temperature, pressure, and vibration signals in multiple time periods according to the denoised signal feature set, and generates a signal feature set;

[0105] First, it is necessary to calculate the fluctuation amplitude, which is the difference between adjacent data points. This helps to reveal the variation range of the signal and the fluctuations in the device state. For example, assume that during a certain device operation, the temperature signal data is: 21.5°C, 22°C, 21.8°C, 21°C, 20.5°C. By calculating the difference between each pair of adjacent data points, the fluctuation amplitude can be obtained. This process is an important basis for subsequent clustering analysis. In addition, by calculating the average value of the cycle, it can help to determine whether the signal shows an upward or downward trend, and this information is crucial for the operating state of the device. For example, if the device temperature gradually increases over each cycle, there is a risk of overheating; if the temperature gradually decreases, it indicates that the device is cooling down. The signal feature set generated through this process provides important data support for subsequent analysis and processing. Using the formula:

[0106] Δx p =|x p,i -x p,i-1 |;

[0107] where, Δx p is the fluctuation amplitude, x p,i is the value of the i-th signal data point, and x p,i-1 is the value of the (i - 1)-th signal data point. Assume the signal data is: x p,1 =21.5°C, x p,2 =22°C, x p,3 =21.8°C, x p,4 =21°C, x p,5 =20.5°C, calculate the fluctuation amplitude:

[0108] Δx p =|22 - 21.5|=0.5°C;

[0109] Δx p =|21.8 - 22|=0.2°C;

[0110] Δx p =|21 - 21.8|=0.8°C;

[0111] Δx p =|20.5 - 21|=0.5°C;

[0112] Through these calculation results, the fluctuation amplitude of the temperature signal is obtained, providing key input data for subsequent clustering analysis.

[0113] The signal clustering analysis sub-module calls the signal feature set, generalizes the signal feature combinations through clustering analysis, extracts the co-occurrence patterns of the signal feature combinations, using the formula:

[0114]

[0115] Calculate the clustering effect, obtain the eigenvalue of the clustering center, and identify the associated feature groups;

[0116] Among them, J Z represents the sum of the squared errors from the signal feature points to their clustering centroids, which is used to evaluate the clustering effect. K is the number of clusters, and N k is the number of signal feature points within the k-th cluster, is the value of the i k -th signal feature in the k-th cluster, μ k is the centroid of the k-th cluster, k is the cluster index, and i k represents the i k -th data point in the k-th cluster;

[0117] The signal clustering analysis sub-module receives the signal feature set generated by the signal feature extraction sub-module and uses the clustering analysis algorithm to classify and analyze the signal features. In the K-means clustering algorithm, the clustering cost function is used to evaluate the clustering effect, and the quality of clustering is measured by calculating the sum of the squares of the distances from the signal feature points to each clustering centroid. This cost function measures the total distance from each data point in the cluster to its cluster center, and the goal is to minimize this cost function to optimize the clustering effect. When using the K-means clustering algorithm, the calculation formula for the clustering cost function is:

[0118]

[0119] Among them, J Z is the clustering cost function, which represents the sum of the squared errors from the signal feature points to the clustering centroids during the clustering process. K is the number of clusters, and N k is the number of signal feature points within the k-th cluster, is the value of the i k -th signal feature in the k-th cluster, and μ k is the centroid of the k-th cluster, representing the eigenvalue of the central feature of the cluster. Assume that the number of clusters K = 2, the number of the first cluster N1 = 3, the number of the second cluster N2 = 2, the centroid of the first cluster μ1 = 22.0, and the centroid of the second cluster μ2 = 21.0. Assume that the signal feature values of the first cluster are The signal feature values of the second cluster are By substituting into the formula to calculate the cost function:

[0120]

[0121] For the cost calculation of the first cluster:

[0122] J Z = ||21.5 - 22.0|| 2+||22.5 - 22.0|| 2 +||21.8 - 22.0|| 2 =(0.5) 2 +(0.5) 2 +(0.2) 2 =0.25 + 0.25 + 0.04 = 0.54;

[0123] Cost calculation for the second cluster:

[0124] J Z =||20.5 - 21.0|| 2 +||21.0 - 21.0|| 2 =(0.5) 2 +(0) 2 =0.25;

[0125] Finally, the clustering cost function J Z =0.54 + 0.25 = 0.79. The smaller the cost function, the better the clustering effect. Continue to minimize the cost function through the optimization process to finally obtain a reasonable clustering result and signal feature grouping.

[0126] The model construction sub - module calls the associated feature grouping, extracts the discrimination rules, constructs the feature structure relationship between device operation data, and generates an operation state classification model;

[0127] Based on the associated feature grouping generated by the signal clustering analysis sub - module, the model construction sub - module extracts and identifies the discrimination rules between signal features. These rules will be used to construct the device operation state classification model. First, based on the signal feature data within each cluster, key indicators related to the device operation state are extracted, such as the rate of temperature change, the fluctuation range of vibration frequency, etc. These features can reflect different working states of the device. For example, in some clusters, if the change amplitude of the temperature signal is large and the pressure signal fluctuates, it represents that the device is in an overload state; while if the vibration signal is strong but the temperature change is small, it represents that the device is in a normal operation state. Through these discrimination rules, the operation state of the device can be divided into multiple categories, such as normal, overload, fault, etc. To improve the classification accuracy, the model construction sub - module analyzes the relationship between features to clearly distinguish different categories of device states and extracts the following discrimination rules: if the temperature fluctuation amplitude is greater than 1°C and the pressure value is greater than the set threshold, then the device is in an overload state. If the vibration signal amplitude is large (greater than the set value) but the temperature change amplitude is small, then the device is in a normal state. By continuously adjusting these discrimination rules and combining with actual operation data, an operation state classification model based on the feature structure relationship is finally generated, which can monitor the operation state of the device in real - time and send an alarm in time when the state changes. This model provides important support for device maintenance and fault prediction.

[0128] Please refer to Figure 4 , the mutation extraction module includes:

[0129] The trend change extraction sub-module calls the operation status classification model, extracts the trend direction change and amplitude change in each period by analyzing the operation data sequence of the device in real time, and generates a trend change feature set;

[0130] The execution process of the trend change extraction sub-module is mainly to extract the trend direction change and amplitude change of each period from the device operation data. First, by monitoring the operation status of the device in real time, the signal data of each period is collected, and based on these data points, the trend change and amplitude change of each period are calculated. For the extraction of the trend direction change, first determine the reference trend of the device signal, and obtain the change amount of the trend direction by comparing the changes of the signals in adjacent periods. At the same time, the amplitude change is obtained by calculating the signal fluctuation amplitude within each period. To calculate the trend direction change, the following formula is used:

[0131] ΔT = |T current -T previous |;

[0132] where, ΔT is the trend direction change, T current is the signal value of the current period, T previous is the signal value of the previous period. Suppose the signal value T current of the current period = 28, and the signal value T previous of the previous period = 26, then calculate the trend direction change:

[0133] ΔT = |28 - 26| = 2;

[0134] The calculation result shows that the trend direction change of the signal in the current period is 2.

[0135] The abnormal period identification sub-module is based on the trend change feature set and uses the formula:

[0136]

[0137] Calculate the abnormal score of each signal period, compare it with the set threshold, identify the abnormal signal period, and obtain the abnormal signal period mark;

[0138] where, S abnormal is the abnormal degree score, m is the total number of data points in the current period, q is the index of the qth data point, X q is the signal value of the qth data point, including temperature, pressure, vibration, T avg is the arithmetic mean of all data points in the current period, ΔT qis the difference between the q-th data point and the previous point, reflecting the instantaneous amplitude change, ΔD q is the trend direction marker of the q-th data point, with values in {-1, 0, 1};

[0139] The abnormal cycle identification sub-module processes and analyzes the temperature values within consecutive sampling cycles based on the temperature signal data collected by the industrial equipment real-time monitoring system. The system collects temperature data at a fixed frequency, and each cycle contains 10 data points. The data points are numbered q = 1 to q = 10 in chronological order. The operating temperature of the special equipment is collected in real-time through a temperature sensor, with a measurement range of 0 - 300 °C. The system automatically calculates the arithmetic mean of all data points within the cycle as the reference temperature, and at the same time calculates the instantaneous temperature change amount between each data point and the previous point, and classifies and marks them according to the change direction. Using the formula:

[0140]

[0141] calculate the temperature anomaly score, where X q is the q-th temperature measurement value, T avg is the cycle average temperature, ΔT q is the adjacent temperature difference, ΔD q is the temperature change direction marker.

[0142] Set T avg to be 184.25 °C within a certain cycle, and the temperature sequence is [180.0, 181.5, 183.2, 182.8, 184.5, 185.1, 183.9, 186.2, 187.0, 188.3]. The sum of the absolute values of the deviations of each data point is calculated to be 19.85 °C, the sum of the squares of the instantaneous changes is 19.86, and the sum of the squares of the direction markers is 10. Substitute the set value for calculation:

[0143]

[0144] S abnormal = 0.1077 + 0.0242 + 0.0172 = 0.1491;

[0145] The calculation results show that the temperature fluctuation in this cycle is within the normal range. The system compares this result with the preset threshold of 0.25 and determines it to be in a normal operating state, ensuring the real-time and reliability of the special equipment operating state monitoring. The system continuously records the anomaly scores of each cycle, and triggers an early warning mechanism when the scores of consecutive multiple cycles exceed the threshold, reminding the operator to conduct inspections and maintenance.

[0146] The mutation mode analysis sub-module calls the abnormal signal cycle markers to analyze the change mode characteristics of the target signal segment, including trend continuity and fluctuation pattern, and generates mutation-driven feature units;

[0147] The task of the mutation pattern analysis sub-module is to identify the change pattern features in the signal cycles marked as abnormal by analyzing them, mainly including trend continuity and fluctuation form. Specifically, first, the signal segments of the abnormal cycles are extracted according to the abnormal cycle markings, and then the change situation of the trend and the fluctuation amplitude within the signal segments are calculated, and the smoothness and volatility of the signals are checked. If the change of the trend direction of the signal is large or the fluctuation amplitude suddenly increases, it indicates that there is a mutation. The mutation-driven feature unit is extracted based on these abnormal fluctuation forms and trend continuities, which can be used as the basis for subsequent fault diagnosis. For example, when the temperature change of the device exceeds the preset threshold, the system will mark this cycle as an abnormal cycle, and then analyze its change pattern. If the temperature changes violently and continuously, the system will generate a mutation-driven feature unit for further analysis of the faults or abnormal behaviors occurring in the device.

[0148] Table 1: Example of Abnormal Cycle Data

[0149] Cycle number Temperature data (°C) Amplitude change Trend direction change Anomaly score 1 25 1.5 0.1 0.28 2 27 1.2 0.2 0.24 3 29 2.5 0.5 0.46 4 30 2.0 0.4 0.45

[0150] As shown in Table 1, the abnormal score of Cycle 3 is 0.46, exceeding the set threshold, so this cycle is marked as an abnormal cycle.

[0151] Please refer to Figure 5 , the state recognition module includes:

[0152] The feature extraction sub-module calls the operating state classification model and the mutation-driven feature unit to extract the trend level, fluctuation amplitude, and change direction from the abnormal signal cycles, and generates a signal feature parameter set;

[0153] The feature extraction sub-module extracts feature information such as trend level, fluctuation amplitude, and change direction from the abnormal signal cycles by calling the operating state classification model and the mutation-driven feature unit, and generates a signal feature parameter set. Specifically, the module first analyzes the operating data of the device in each cycle, extracts the trend direction (for example, the trend of the device load gradually increasing or decreasing), and then calculates the fluctuation amplitude within this cycle (the fluctuation range of the device operating state). At the same time, by analyzing the change direction of the device signal within the cycle, it is determined whether the device shows an abnormal fluctuation pattern. For example, if the load fluctuation of the device in a certain cycle is greater than the preset threshold and the fluctuation direction changes suddenly, then this cycle is considered an abnormal cycle. Finally, these extracted feature data will be used as the basic parameters for subsequent similarity calculation to form a signal feature parameter set.

[0154] The similarity calculation sub-module is based on the signal feature parameters. According to the trend level, amplitude value, and direction change amount, through comparative analysis with the data features in multiple associated feature groups, using the formula:

[0155]

[0156] Calculate the similarity value and obtain the feature matching similarity value;

[0157] Among them, S m′ is the feature matching similarity value, M is the total number of features, w p is the weight coefficient of the p-th feature, X m′p is the p-th eigenvalue of the signal to be recognized, X p is the p-th feature reference value in the known associated feature group, σ m′ is the variance value of the fluctuation of the p-th feature of the signal to be recognized, σ p is the variance value of the fluctuation of the p-th feature in the known associated feature group, α is an adjustment factor, p represents the index of the feature, and m' represents the index of the feature group of the signal to be recognized;

[0158] According to the extracted signal feature parameters, the similarity calculation sub-module compares and analyzes the data features in multiple associated feature groups through the trend level, amplitude value, and direction change amount, and uses the following formula to calculate the similarity value:

[0159]

[0160] Among them, S m′ is the feature matching similarity value, M is the total number of features, w p is the weight coefficient of the p-th feature, X m′p is the p-th eigenvalue of the signal to be recognized, X p is the p-th feature reference value in the known associated feature group, σ m′ is the variance value of the fluctuation of the p-th feature of the signal to be recognized, σ p is the variance value of the fluctuation of the p-th feature in the known associated feature group, α is an adjustment factor, represents the index of the feature, and p represents the index of the feature group of the signal to be recognized. During the calculation process, the similarity is calculated by comparing the differences between the signal to be recognized and each feature in the known feature group. Assume that the value of the signal to be recognized on a certain feature is X m′p = 0.8, the known feature reference value is X p = 0.75, the variance of the fluctuation of this feature is σ m′ = 0.2, the known reference variance of the fluctuation is σ p = 0.1, and the weight coefficient of this feature is w p = 0.5, the adjustment factor α = 0.05. Substitute into the formula to calculate the similarity value:

[0161]

[0162] According to the calculation results, the similarity value is 0.055. This means that the similarity between the signal to be recognized and the known feature group is low, indicating that the device is abnormal or has a large difference from the historical data. This similarity value will be further used for state classification judgment.

[0163] The state classification sub-module, based on the feature matching similarity value, corresponds with the state labels in real time to identify the operating state of the device and generate a state feature classification label.

[0164] Compare the calculated similarity value with multiple preset device state labels, such as "normal", "fault", or "warning" states. By setting a threshold to judge the level of similarity, if the similarity value is higher than a certain preset value, the device is classified as the "normal" state; if the similarity value is lower than a certain threshold, the device is determined to be in the "fault" or "warning" state. For example, if the feature matching similarity value of the signal to be recognized is 0.055, and the known similarity threshold for the "normal" state is 0.8, and the similarity threshold for the "fault" state is 0.3, then the signal to be recognized will be determined to be in the "fault" state because its similarity value is low. Finally, this process will output a state label for the device operation, providing a basis for real-time monitoring for subsequent maintenance or decision-making.

[0165] Please refer to Figure 6 , the trend prediction module includes:

[0166] The trend extraction sub-module obtains the denoised signal feature set and the state feature classification label, judges the direction consistency of the signal in each period, calculates the amplitude difference, and counts the trend maintenance time, marks the trend direction, amplitude ratio, and trend persistence, and obtains the periodic trend feature set.

[0167] First, obtain the denoised signal feature set, including real-time data provided by multiple sensors, such as parameters like temperature, pressure, vibration, etc. Assume that temperature and vibration data are used, and the sensors record the values of these signals at different time points. These data are first denoised through a filtering algorithm to remove the high-frequency noise. Next, judge the direction consistency of the signal in each period. Assume that the temperature of a signal sequence shows an upward trend at 5 consecutive moments. The calculation of the amplitude difference indicates how much the signal changes, and further judge whether this signal change conforms to the normal operation mode of the device. For the amplitude difference, assume that the temperature change in one period is 0.2 °C and in another period is 0.4 °C, then the amplitude increase is 0.2 °C. Subsequently, count the trend maintenance time of the signal. For example, if the vibration signal of a device changes relatively little in 3 consecutive periods, it is considered to have a high persistence. Through this process, finally, mark the trend direction, amplitude ratio, and trend persistence of the signal in each period, and generate the periodic trend feature set.

[0168] Formula:

[0169] ΔT i′ = T i′ - T i′-1 ;

[0170] Wherein, ΔT i′ represents the temperature change value of the i'-th cycle, T i′ is the temperature value of the i'-th cycle, and T i′-1 is the temperature value of the previous cycle.

[0171] It is set that the temperature of the device in cycle 1 is 22 °C, the temperature in cycle 2 is 22.4 °C, and the temperature in cycle 3 is 22.6 °C. In cycle 2, relative to cycle 1, the temperature change is:

[0172] T2 - T1 = 22.4 °C - 22 °C = 0.4 °C.

[0173] In cycle 3, relative to cycle 2, the temperature change is:

[0174] T3 - T2 = 22.6 °C - 22.4 °C = 0.2 °C. Thus, the calculation results of the increase difference are 0.4 °C and 0.2 °C, indicating that there is a certain trend fluctuation in the temperature change of the device.

[0175] The probability analysis sub-module, according to the cycle trend feature set, adopts the formula:

[0176]

[0177] to calculate the occurrence probability of the abnormal state in each cycle and generate a cycle abnormal probability sequence;

[0178] Wherein, P' i′ represents the occurrence probability of the abnormal state in the i'-th cycle, TQ i′ represents the trend direction mark in the i'-th cycle, SZ i′ represents the increase ratio in the i'-th cycle, DB i′ represents the fluctuation degree value in the i'-th cycle, VZ i′ represents the vibration amplitude in the i'-th cycle, RK i′ represents the trend duration in the i'-th cycle, and i' is the cycle number;

[0179] In the probability analysis sub-module, first, according to the set of periodic trend features generated by the previous module, specific feature information for each period is obtained, such as trend direction markers, increase ratios, fluctuation degrees, and vibration amplitudes. Suppose that in a certain period, the trend direction marker is upward (TQ = 1), the increase ratio (SZ = 0.4), the fluctuation degree (DB = 0.3), the vibration amplitude (VZ = 0.1), and the trend duration (RK = 3). These parameters are substituted into the formula for calculation to obtain the probability P' of the occurrence of an abnormal state in this period. i′ , the formula is as follows:

[0180]

[0181] Set TQ i′ = 1 - SZ i′ = 0.4 - DB i′ = 0.3 - VZ i′ = 0.1 - RK i′ = 3, substitute into the formula:

[0182]

[0183] Thus, the calculated probability P' of the occurrence of an abnormal state i′ = 4, indicating that the equipment state in this period is abnormal.

[0184] The node positioning sub-module calls the periodic abnormal probability sequence, analyzes the change trend of the abnormal probability in each period, detects the mutation interval of the abnormal probability, predicts the time nodes of the state change section, and obtains the trend-driven warning nodes;

[0185] In the node positioning sub-module, the periodic abnormal probability sequence is called to analyze the change trend of the abnormal probability in each period. If the abnormal probability in a certain period exceeds the set threshold (for example, the set threshold is 3), it is considered that there is an abnormality in this period. On this basis, by analyzing the mutation interval of the abnormal probability, the time nodes of the state changes that occur to the equipment are speculated. For example, in the abnormal probability sequence of a certain equipment, the abnormal probability between the 10th period and the 12th period rises sharply from 3 to 7, indicating that the equipment fails in the 12th period. Therefore, the predicted time node of the equipment state change is the 12th period. By analyzing the mutations in multiple periods, the specific time period from the normal state to the abnormal state is finally located, and the predicted state change prediction node is obtained. Formula:

[0186] ΔP' i′ = P' i′ - P' i′-1 ;

[0187] Among them, ΔP' i′ represents the change amount of the abnormal state occurrence probability in the i'-th period, P'i′ is the probability of occurrence of an abnormal state in cycle i′, P′ i′-1 is the probability of occurrence of an abnormal state in cycle i′ - 1.

[0188] Set the abnormal probability in the 10th cycle to 3 and the abnormal probability in the 12th cycle to 7, and substitute into the formula:

[0189] ΔP′ i′ = 7 - 3 = 4;

[0190] This indicates that from the 10th cycle to the 12th cycle, the abnormal probability has increased by 4, indicating that the device state has changed significantly. Based on this information, it can be predicted that the probability of the device failing in the 12th cycle is relatively high, and the time node of the state change is the 12th cycle.

[0191] Table 2: Example of cycle data

[0192]

[0193] As shown in Table 2, it shows the data of different cycles and the calculated probability of occurrence of an abnormal state P′ i′ . The characteristic data of each cycle is the real-time data collected by the device sensor. The trend direction mark, increase rate, fluctuation degree, vibration amplitude, and trend duration of the cycle are all calculated according to the formula, so as to obtain the probability of occurrence of an abnormal state.

[0194] A real-time detection method for the state of special equipment, the real-time detection method for the state of special equipment is executed based on the above-mentioned real-time detection system for the state of special equipment, and includes the following steps:

[0195] S1: Based on the sensor monitoring data, obtain the temperature, pressure, and vibration signals during the operation of the special equipment. By analyzing the continuity of each signal change, detect and eliminate abnormal jump signals, denoise the data, and generate a denoised signal feature set;

[0196] S2: Based on the denoised signal feature set, extract the data features of each type of signal in multiple time cycles, combine clustering analysis to summarize the signal feature combinations, identify the co-occurrence patterns between features, classify relevant features, and extract discrimination rules to generate an operating state classification model;

[0197] S3: Based on the operating state classification model, by obtaining the real-time operating data sequence of the device, analyze the trend direction change and amplitude change in each cycle, extract abnormal change cycles, and analyze the change pattern of the signal in the target cycle, including trend continuity and fluctuation form, to generate a mutation-driven feature unit;

[0198] S4: Based on the operating state classification model and the mutation-driven feature unit, extract the data features in the abnormal signal, and conduct a comparative analysis with the associated feature groups to identify the current operating state of the device and generate a status feature classification label;

[0199] S5: Based on the denoised signal feature set and the status feature classification label, extract the trend direction marker, the amplitude increase ratio sequence, and the trend persistence of each type of signal within multiple cycles. By analyzing the signal change trend, calculate the probability of the occurrence of the abnormal state in each cycle, predict the time node of the state change, and obtain the trend-driven warning node.

[0200] The above is only a preferred embodiment of the present invention, and it does not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A real-time detection system for the state of special equipment, characterized in that, The system includes: The data processing module obtains the sensor monitoring data, including temperature signals, pressure signals, and vibration signals. By analyzing the continuity of data changes, it detects and eliminates abnormal jump signals, denoises the data, and generates a denoised signal feature set. The association construction module extracts the data features of each type of signal in multiple time periods according to the denoised signal feature set, induces the co-occurrence patterns of signal feature combinations through cluster analysis, identifies associated feature groups, extracts discrimination rules, constructs the feature structure relationship between operation data, and generates an operation state classification model. The mutation extraction module calls the operation state classification model, analyzes the trend direction change and amplitude change in each period by real-time analyzing the operation data sequence of the device, identifies the signal periods with abnormal changes, and analyzes the change mode features of the target signal segment, including trend continuity and fluctuation form, to generate mutation-driven feature units. The state recognition module calls the operation state classification model and the mutation-driven feature units, extracts the data features of abnormal signals, and identifies the operation state of the device by comparing with the data features of multiple associated feature groups, generating state feature classification labels.

2. The special equipment status real-time detection system according to claim 1, characterized in that, The denoised signal feature set includes signal change trend labels, cycle stability identifiers, and abnormal segment markers. The operation state classification model includes a signal feature grouping structure, a state clustering label group, and a classification discrimination factor set. The mutation-driven feature units are specifically trend reversal nodes, amplitude abnormal sections, and change structure feature blocks. The state feature classification labels include state type numbers, dominant signal feature items, and linked signal classification items.

3. The real-time detection system for the state of special equipment according to claim 1, wherein The data processing module includes: The data acquisition sub-module obtains the sensor monitoring data, collects temperature signals, pressure signals, and vibration signals during the operation of special equipment, and generates original equipment operation data. The signal extraction sub-module calls the original equipment operation data, extracts the fluctuation amplitude, cycle average change direction, and extreme value difference of each type of signal, and generates a signal feature data set. The data denoising sub-module calls the signal feature data set, analyzes the continuity of data changes, and uses the formula: Calculate the jump sensitivity coefficient, filter the abnormal jump points of the signal by comparing the coefficient with the jump determination threshold, and obtain the denoised signal feature set. Among them, D jump is the jump sensitivity coefficient, N is the total number of signal data points, x i is the value of the i-th signal data point, x i-1 is the value of the (i - 1)-th signal data point, μ is the mean of the signal data set, i is the index of the current signal data point for calculating the difference between adjacent data points, i - 1 is the index of the data point before the current signal data point i, j is the index in the signal data set for calculating the variance of the signal data, x j is the value of the j-th signal data point.

4. The real-time detection system for the status of special equipment according to claim 3, wherein, The association construction module includes: The signal feature extraction sub-module extracts the trend level, fluctuation amplitude, and change direction of temperature, pressure, and vibration signals in multiple time periods according to the denoised signal feature set, and generates a signal feature set. The signal cluster analysis sub-module calls the signal feature set, induces the signal feature combinations through cluster analysis, extracts the co-occurrence patterns of signal feature combinations, and uses the formula: Calculate the clustering effect, obtain the clustering center feature value, and identify associated feature groups. Among them, J Z represents the sum of squared errors from signal feature points to their clustering centroids, used to evaluate the clustering effect, K is the number of clusters, N k is the number of signal feature points within the k-th cluster, is the value of the i k -th signal feature in the k-th cluster, μ k is the centroid of the k-th cluster, k is the cluster index, i k represents the i k -th data point in the k-th cluster; The model construction sub-module calls the associated feature groups, extracts discrimination rules, constructs the feature structure relationship between device operation data, and generates an operation state classification model.

5. The special equipment status real-time detection system according to claim 4, characterized in that, The mutation extraction module includes: The trend change extraction sub-module calls the operating status classification model, extracts the trend direction change and amplitude change in each period by analyzing the real-time operating data sequence of the device, and generates a trend change feature set; The abnormal period identification sub-module is based on the trend change feature set and uses the formula: Calculate the abnormal score of each signal period, compare it with the set threshold, identify the abnormal signal period, and obtain the abnormal signal period label; Among them, S abnormal is the anomaly degree score, m is the total number of data points in the current period, q is the index of the q-th data point, and X q is the signal value of the q-th data point, including temperature, pressure, and vibration. T avg is the arithmetic mean of all data points in the current period, and ΔT q is the difference between the q-th data point and the previous point, reflecting the instantaneous amplitude change. ΔD q is the trend direction marker of the q-th data point; The mutation mode analysis sub-module calls the abnormal signal period label, analyzes the change mode characteristics of the target signal segment, including trend continuity and fluctuation form, and generates a mutation-driven feature unit.

6. The real-time special equipment status detection system according to claim 5, wherein, The state recognition module includes: The feature extraction sub-module calls the operating status classification model and the mutation-driven feature unit, extracts the trend level, fluctuation amplitude, and change direction from the abnormal signal period, and generates a signal feature parameter set; The similarity calculation sub-module is based on the signal feature parameters, according to the trend level, amplitude value, and direction change amount, and through comparative analysis with the data features in multiple associated feature groups, uses the formula: Calculate the similarity value and obtain the feature matching similarity value; Among them, S m′ is the feature matching similarity value, M is the total number of features, w p is the weight coefficient of the p-th feature, X m′p is the p-th eigenvalue of the signal to be recognized, X p is the p-th feature reference value in the known associated feature group, σ m′ is the variance value of the p-th feature of the signal to be recognized, σ p is the variance value of the p-th feature in the known associated feature group, α is an adjustment factor, p represents the index of the feature, and m′ represents the feature group index of the signal to be recognized; The state classification sub-module, according to the feature matching similarity value, corresponds to the state label in real time to identify the operating state of the device and generate a state feature classification label.

7. The real-time special equipment status detection system according to claim 1, characterized in that, The system further includes: The trend prediction module extracts the trend direction label, increase ratio sequence, and trend persistence in multiple periods according to the denoised signal feature set and the state feature classification label, analyzes the signal change trend, calculates the probability of abnormal state occurrence, and predicts the time node of the state change section to obtain a trend-driven early warning node; The trend-driven early warning node includes a state change start index, a trend continuation flag bit, and an early warning level parameter.

8. The real-time special equipment status detection system according to claim 7, characterized in that The trend prediction module includes: The trend extraction sub-module obtains the denoised signal feature set and the state feature classification label, judges the direction consistency of the signal in each period, calculates the increase difference, and counts the trend maintenance time, marks the trend direction, increase ratio, and trend persistence, and obtains a periodic trend feature set; The probability analysis sub-module is based on the periodic trend feature set and uses the formula: Calculate the probability of abnormal state occurrence in each period and generate a periodic abnormal probability sequence; Among them, P' i′ represents the abnormal state occurrence probability of the i'-th period, TQ i′ represents the trend direction mark of the i'-th period, SZ i′ represents the increase rate of the i'-th period, DB i′ represents the fluctuation degree value of the i'-th period, VZ i′ represents the vibration amplitude of the i'-th period, RK i′ represents the trend duration of the i'-th period, where i' is the period number; The node positioning sub-module calls the periodic abnormal probability sequence, analyzes the change trend of the abnormal probability in each period, detects the mutation interval of the abnormal probability, predicts the time node of the state change section, and obtains a trend-driven early warning node.

9. A real-time detection method for the state of special equipment, characterized in that, Execute according to the special equipment state real-time detection system described in any one of claims 1-8, including the following steps: S1: Based on the sensor monitoring data, obtain the temperature, pressure, and vibration signals during the operation of the special equipment, detect and eliminate abnormal jump signals by analyzing the continuity of each signal change, denoise the data, and generate a denoised signal feature set; S2: Based on the denoised signal feature set, extract the data features of each type of signal within multiple time periods, combine cluster analysis to summarize the signal feature combinations, identify the co-occurrence patterns between features, classify relevant features and extract discrimination rules, and generate an operating state classification model; S3: Based on the operating state classification model, by obtaining the real-time operating data sequence of the device, analyze the trend direction change and amplitude change in each period, extract the abnormal change period, and analyze the change pattern of the signal in the target period, including trend continuity and fluctuation form, to generate a mutation-driven feature unit; S4: Based on the operating state classification model and the mutation-driven feature unit, extract the data features in the abnormal signal, and conduct a comparative analysis with the associated feature groups to identify the current operating state of the device and generate a state feature classification label; S5: Based on the denoised signal feature set and the state feature classification label, extract the trend direction markers, amplitude increase ratio sequences, and trend persistence of each type of signal within multiple periods. By analyzing the signal change trend, calculate the probability of abnormal state occurrence in each period, predict the time node of state change, and obtain the trend-driven early warning node.

Images