Phase selection closing device automatic detection system and method based on intelligent algorithm

Abstract

The invention relates to the technical field of state monitoring and fault diagnosis, in particular to an automatic detection system and method for a phase selection closing device based on an intelligent algorithm, and the method comprises the steps: a multi-mode sensing module applies an active excitation signal and obtains an active mechanical response signal when the device is in a standby state, and obtains a passive operation signal when the device is in a closing operation; the feature extraction module extracts active mechanical features and passive dynamic features respectively, and the active mechanical features and the passive dynamic features are spliced into target feature vectors; the deviation prediction module establishes an explicit degradation equation through a symbolic regression algorithm, inputs a target feature vector to obtain predicted closing time deviation and outputs abnormal early warning; and after receiving the early warning, the diagnostic analysis module deduces a device fault source based on a target feature vector and a Bayesian causal network. According to the invention, active harmonic detection is adopted when the device is in a standby state, and passive dynamic monitoring is carried out when the device is switched on, so that effective separation of pure mechanical characteristics and electromechanical coupling dynamic characteristics of equipment is realized.

Description

Technical Field

[0001] This invention relates to the field of condition monitoring and fault diagnosis technology, and in particular to an automatic detection system and method for a phase selection and closing device based on intelligent algorithms. Background Technology

[0002] In modern high-voltage and ultra-high-voltage power grids, transient inrush currents and operational overvoltages generated during circuit breaker closing operations are major factors threatening grid safety and damaging the lifespan of expensive equipment such as transformers. The phase-selective closing device is a key control unit designed to address this problem. Its core function is to precisely control the circuit breaker to close its contacts at a specific voltage phase, thereby minimizing the aforementioned transient impacts. However, this high-precision control is highly dependent on the long-term stability of the circuit breaker's operating mechanism's mechanical characteristics. As the equipment operates, inevitable wear, loosening, or lubrication aging of mechanical components will continuously cause drift in its closing characteristics. Once this drift exceeds the allowable range, the phase-selective function will fail, seriously threatening the safe and stable operation of the power system.

[0003] Existing detection technologies primarily focus on monitoring the final result of closing time deviation. These technologies calculate the final closing time deviation by comparing the expected closing command time issued by the system with the actual contact closing time sensed by sensors during the closing operation. The advantage of this method is that it provides an intuitive and quantitative indicator for directly verifying whether the device's terminal performance meets the accuracy thresholds required by the operating procedures. When equipment performance deteriorates and the time deviation exceeds the standard, this type of method can reliably trigger a "fault" alarm, providing the operating unit with a clear "qualified / unqualified" judgment basis.

[0004] However, the aforementioned existing technologies have significant limitations in practical applications. First, this monitoring method, which relies solely on the final time deviation, is essentially a reactive measure. It can only respond passively when equipment performance has deteriorated to a severe degree, resulting in a clear deviation exceeding the limit. It lacks the ability to detect early, subtle signs of performance degradation, thus failing to provide proactive condition warnings. Second, when an alarm occurs, this method cannot pinpoint the root cause of the deviation exceeding the limit. The phase-selective closing device is a complex electromechanical coupling system. The observed time deviation may be caused by pure mechanical wear and jamming, or by electrical delays in the control circuit or coil aging. The signals acquired by existing technologies during the closing process are highly coupled, making it impossible to accurately distinguish between mechanical and electrical faults, i.e., lacking precise root cause diagnosis capabilities. Furthermore, even with the introduction of complex algorithms, its black-box nature prevents maintenance personnel from understanding the specific mechanisms of degradation, leading to blind maintenance decisions. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the difficulties in achieving early warning of degradation and the inability to accurately distinguish the root causes of mechanical and electrical faults in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention provides an automatic detection system for a phase selection and closing device based on an intelligent algorithm, comprising: The multimodal sensing module is used to apply an active excitation signal and acquire an active mechanical response signal when the phase selection closing device is in standby mode; and to acquire a passive operation signal when the phase selection closing device is in closing operation. The feature extraction module, connected to the multimodal perception module, is used to extract active mechanical features from the active excitation signal and the active mechanical response signal, extract passive dynamic features from the passive operation signal, and concatenate the active mechanical features and the passive dynamic features into a target feature vector. The deviation prediction module, connected to the feature extraction module, is used to establish an explicit degradation equation representing the relationship between the target feature vector and the closing time deviation using a symbolic regression algorithm, input the target feature vector into the explicit degradation equation to obtain the predicted closing time deviation, and output an abnormal warning based on the predicted closing time deviation. The diagnostic analysis module, connected to the feature extraction module and the deviation prediction module, is used to infer the root cause of the fault in the phase selection and closing device based on the target feature vector and using a Bayesian causal network when the deviation prediction module outputs an abnormal warning.

[0007] Preferably, the multimodal sensing module includes: The active excitation unit, including a piezoelectric ceramic exciter, is used to apply an active excitation signal to the mechanical structure when the phase selection closing device is in standby mode. The response pickup unit includes a wideband vibration sensor, which is used to acquire the active mechanical response signal generated by the active excitation signal of the mechanical structure when the phase selection closing device is in standby mode.

[0008] Preferably, the multimodal sensing module further includes: The environmental sensing unit is used to acquire environmental parameters of the environment in which the phase selection and closing device is located.

[0009] Preferably, the active mechanical features are extracted from the active excitation signal and the active mechanical response signal, by means of: Fourier transforms are performed on the active excitation signal and the active mechanical response signal respectively, and the ratio of the transformed active mechanical response signal to the transformed active excitation signal is used as the mechanical transfer function of the phase selection and closing device in the frequency domain. Based on the mechanical transfer function of the baseline state, obtain the baseline mechanical modal parameters; Based on the mechanical transfer function of the current detection cycle, obtain the mechanical modal parameters of the current detection cycle; The difference between the mechanical modal parameters of the current detection cycle and the corresponding baseline mechanical modal parameters is used as the active mechanical feature.

[0010] Preferably, the passive operation signal includes an acoustic timing signal, a vibration timing signal, and a coil current timing signal.

[0011] Preferably, the passive dynamic features extracted by the feature extraction module include: Based on nonlinear dynamics analysis, recursive graph quantization features or fractal dimension features are extracted from the acoustic time-series signal or vibration time-series signal.

[0012] Preferably, the passive dynamic features extracted by the feature extraction module further include: The coil response time characteristics or coil drive energy characteristics extracted from the coil current timing signal.

[0013] Preferably, the deviation prediction module outputs an abnormal warning based on the predicted closing time deviation, including: If the predicted closing time deviation is greater than the preset warning threshold but less than the preset fault threshold, the first abnormal warning is output to indicate that the phase selection closing device has deviated from the optimal health state. Linear regression analysis is performed on the predicted closing time deviation of multiple consecutive detection cycles to calculate the slope; if the slope is greater than the preset alarm slope, a second abnormal alarm is output to indicate that the phase selection closing device is in a deterioration trend.

[0014] Preferably, the nodes of the Bayesian causal network utilized by the diagnostic analysis module include: The root node represents the fault hypothesis, including mechanical faults, electrical faults, and sensor faults; The intermediate node representing characteristic evidence includes the active mechanical feature and the passive dynamic feature; Leaf node representing closing time deviation.

[0015] This invention also provides an automatic detection method for a phase selection and closing device based on an intelligent algorithm, comprising: When the phase selection closing device is in standby mode, an active excitation signal is applied and an active mechanical response signal is acquired; and when the phase selection closing device is in closing operation, a passive operation signal is acquired. Active mechanical features are extracted from the active excitation signal and the active mechanical response signal, and passive dynamic features are extracted from the passive operation signal. The active mechanical features and the passive dynamic features are then concatenated into a target feature vector. After establishing an explicit degradation equation representing the relationship between the target feature vector and the closing time deviation using the symbolic regression algorithm, the target feature vector is input into the explicit degradation equation to obtain the predicted closing time deviation, and an abnormal warning is output based on the predicted closing time deviation. When the deviation prediction module outputs an abnormal warning, the root cause of the fault in the phase selection and closing device is inferred based on the target feature vector and using a Bayesian causal network.

[0016] Compared with the prior art, the above-described technical solution of the present invention has the following advantages: The automatic detection system for phase selection and closing devices based on intelligent algorithms described in this invention achieves effective separation of the purely mechanical characteristics and electromechanical coupling dynamic characteristics of the equipment by employing active harmonic detection when the device is in standby mode and passive dynamic monitoring when it is closed. This active-passive combined data acquisition strategy overcomes the problem of confusion between mechanical signals and electromagnetic interference signals in traditional methods, providing a high-confidence, decoupled data foundation for subsequent accurate fault root cause diagnosis.

[0017] Furthermore, by introducing nonlinear dynamic analysis to process the passively monitored vibration and acoustic signals, this invention can capture the subtle dynamic changes in the early stages of the device's evolution from health to failure. Compared with traditional time-frequency domain statistical features, these nonlinear features are more sensitive to the reduction of motion determinism or changes in system complexity, thereby improving the system's ability to perceive early potential failures and achieving earlier forward-looking warnings.

[0018] Furthermore, this invention utilizes explicit degradation equations for real-time deviation prediction and establishes a dual-mode early warning mechanism combining trend warning and threshold warning, thereby achieving proactive management of equipment status. The system can identify the continuous degradation trend or slight deviation of performance in advance before the actual closing time deviation exceeds the qualified range, thus transforming traditional post-fault maintenance into status-based predictive maintenance, effectively avoiding unexpected downtime and ensuring the operational reliability of the phase selection closing device. Attached Figure Description

[0019] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a structural diagram of an automatic detection system for a phase selection and closing device based on an intelligent algorithm, according to the present invention. Figure 2 This is a flowchart of an automatic detection method for a phase selection and closing device based on an intelligent algorithm, according to the present invention. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0021] Reference Figure 1 As shown, the present invention provides an automatic detection system for a phase selection and closing device based on an intelligent algorithm, comprising: The multimodal sensing module is used to apply an active excitation signal and acquire an active mechanical response signal when the phase selection closing device is in standby mode; and to acquire a passive operation signal when the phase selection closing device is in closing operation. The feature extraction module, connected to the multimodal perception module, is used to extract active mechanical features from the active excitation signal and the active mechanical response signal, extract passive dynamic features from the passive operation signal, and concatenate the active mechanical features and the passive dynamic features into a target feature vector. The deviation prediction module, connected to the feature extraction module, is used to establish an explicit degradation equation representing the relationship between the target feature vector and the closing time deviation using a symbolic regression algorithm, input the target feature vector into the explicit degradation equation to obtain the predicted closing time deviation, and output an abnormal warning based on the predicted closing time deviation. The diagnostic analysis module, connected to the feature extraction module and the deviation prediction module, is used to infer the root cause of the fault in the phase selection and closing device based on the target feature vector and using a Bayesian causal network when the deviation prediction module outputs an abnormal warning.

[0022] The multimodal sensing module, serving as the fundamental data source for the automatic detection system of this invention, aims to comprehensively and three-dimensionally capture the physical information of the phase selection and closing device under different states through a combined active and passive sensing strategy. This module's design overcomes the limitations of traditional detection methods that rely solely on single operational results, aiming to uncover deeper characteristics of equipment health from process dynamics.

[0023] The multimodal sensing module's functionality can be divided into two mutually cooperating operating modes: an active harmonic detection mode in standby mode and a passive dynamic monitoring mode during the closing operation. This dual-modal design aims to effectively separate the purely mechanical structural characteristics of the equipment from the electromechanical coupling dynamic characteristics during actual operation, providing high-reliability, low-coupling data input for subsequent feature extraction and accurate diagnosis.

[0024] To achieve active harmonic detection, the multimodal sensing module integrates an active excitation unit and a high-sensitivity response pickup unit.

[0025] The active excitation unit can be a piezoelectric ceramic exciter, which is fixed to a non-moving part such as the mechanical mechanism housing of the device, and is used to apply a controllable, preset active excitation signal to the mechanical structure when the phase selection and closing device is in standby mode.

[0026] In this embodiment, the active excitation signal is designed as a wideband sweep frequency signal, denoted as . The purpose of using this signal format is that its frequency can be maintained within a wide frequency range over a preset time period. Inner linear or logarithmic sweep. The technical advantage of this design is that it can continuously excite multiple natural frequency points that may exist in the mechanical structure of the device, thereby comprehensively exploring its vibration response characteristics across the entire frequency band, rather than just the response at a single frequency. This provides sufficient excitation conditions for subsequent calculation of the complete mechanical transfer function and capture of subtle structural changes.

[0027] Corresponding to the active excitation, the response pickup unit uses a wideband vibration sensor to collect the active mechanical response signal generated by the active excitation signal of the mechanical structure when the phase selection closing device is in standby mode.

[0028] When the piezoelectric exciter applies When a signal is received, the vibration sensor synchronously acquires the vibration response signal generated by the mechanical structure, which is recorded as the active mechanical response signal. This active detection process is deliberately performed when the device is online but in standby mode. The key reason is that in this state, there is no strong electromagnetic interference or severe mechanical impact associated with closing the circuit breaker, thus the collected signals are more accurate. It can purely reflect the transmission characteristics of the mechanical structure itself, thereby establishing a stable and clean mechanical health fingerprint baseline.

[0029] The collected active excitation signal With active mechanical response signal It was used in subsequent calculations of the device's mechanical transfer function. The calculation formula is as follows: ; in, This represents the mechanical transfer function of the device in the frequency domain; For frequency; It is a wideband sweep frequency active excitation signal in the time domain; The active mechanical response signal in the time domain; This represents the Fourier transform that converts a time-domain signal into a frequency-domain signal. This forms the direct basis for the subsequent extraction of active mechanical features.

[0030] To achieve passive dynamic monitoring, the multimodal sensing module also integrates sensing units for capturing the complete dynamic process of the device under the actual closing command. These sensing units include, but are not limited to, high-frequency acoustic sensors, wideband vibration sensors, and high-sampling-rate coil current sensors. These sensors are configured to be synchronously triggered the instant the system receives the closing command, recording the entire action process with a time resolution of milliseconds or microseconds.

[0031] Based on the sensing unit, the passive operation signals collected include acoustic timing signals, vibration timing signals, and coil current timing signals.

[0032] Specifically, acoustic and vibration sensors, in passive monitoring mode, are used to collect acoustic timing signals generated throughout the entire closing process. and vibration timing signal These signals contain rich process information from the unlocking of the drive mechanism, the engagement of the electromagnet, the linkage transmission, to the final contact impact and closure. Compared to focusing only on the final result, the waveform, amplitude, and spectral characteristics of these process signals can reflect abnormalities in the mechanical motion process earlier and more intuitively, such as increased friction, changes in impact intensity, or abnormal vibrations.

[0033] The coil current sensor features a high sampling rate. Its purpose is not only to determine whether the closing coil is energized, but also to precisely capture the complete waveform profile of the coil current, recording it as the coil current timing signal. The current profile contains rich electromechanical coupling information; the slope of the current rising segment reflects the electrical health of the control circuit and the coil itself; inflection points or plateaus on the current curve are usually directly related to the motion state of the electromagnet armature. Therefore, for The precise capture of data is crucial for distinguishing between electrical faults and root causes such as mechanical jamming.

[0034] To enhance the robustness and environmental adaptability of the diagnostic model, the multimodal sensing module can also integrate an environmental sensing unit to acquire environmental parameters of the environment in which the phase selection and closing device is located, such as temperature. and humidity These environmental parameters, along with active and passive signals, will be recorded. When subsequent degradation model analysis detects drift in equipment performance parameters, environmental factors can be introduced as auxiliary variables to effectively distinguish between irreversible degradation of the equipment itself and performance fluctuations caused solely by environmental changes, thus avoiding potential false alarms.

[0035] The multimodal sensing module ensures that all active and passive signals, time-stamped information, and environmental parameters have a unified time reference during acquisition through a unified data acquisition and synchronization control unit. This high-precision data synchronization is crucial for subsequent analysis of the temporal relationships and causal correlations between signals. Finally, the module transmits the packaged, precisely time-stamped multidimensional data stream to the subsequent feature extraction module for in-depth processing.

[0036] The feature extraction module, serving as the core hub connecting the multimodal perception module and subsequent modeling and diagnostic modules, bears the crucial task of transforming raw, multidimensional, and heterogeneous perception data into structured, information-dense feature vectors with clear physical meaning. The design philosophy of this module is to maximize the extraction of depth information representing purely mechanical properties and electromechanical coupling dynamics from both active detection signals and passive monitoring signals.

[0037] The feature extraction module first processes the active harmonic detection data collected by the multimodal sensing module to extract active mechanical features.

[0038] Specifically, the feature extraction module receives active excitation signals acquired during the device's standby state. and its corresponding active mechanical response signal The active mechanical features are extracted from them using the following method: Fourier transforms are performed on the active excitation signal and the active mechanical response signal respectively, and the ratio of the transformed active mechanical response signal to the transformed active excitation signal is used as the mechanical transfer function of the phase selection and closing device in the frequency domain. Based on the mechanical transfer function of the baseline state, obtain the baseline mechanical modal parameters; Based on the mechanical transfer function of the current detection cycle, obtain the mechanical modal parameters of the current detection cycle; The difference between the mechanical modal parameters of the current detection cycle and the corresponding baseline mechanical modal parameters is used as the active mechanical feature.

[0039] Specifically, after receiving the active excitation signal and active mechanical response signal for the current detection cycle, the mechanical transfer function for the current detection cycle is calculated using the following formula: ; Where k represents the detection period index, Let be the mechanical transfer function of the phase selection and closing device in the frequency domain during the k-th detection cycle. This represents the active mechanical response signal in the k-th detection cycle. This is the active excitation signal for the k-th detection cycle.

[0040] To achieve drift-based diagnosis, the feature extraction module also needs to load or establish the mechanical transfer function of the healthy baseline state during system initialization, denoted as... This baseline was measured using the same active harmonic detection method when the device was confirmed to be in a healthy state. Furthermore, from... and In this study, a set of key mechanical modal parameters were extracted. These mechanical modal parameters include, but are not limited to, the resonant frequencies of each mode. and the corresponding damping ratio The purpose of this step is to reduce the complexity of the frequency response curve to a few core physical parameters.

[0041] In this embodiment, the final active mechanical feature Mechanical modal parameters defined as those of the current detection cycle Relative to baseline mechanical modal parameters Drift amount: ; ; in, For the detection period index, for example, the first... Second test; The first part representing the mechanical structure First mode; Indicates the first During the first testing cycle, the measured mechanical structure Damping ratio of the first mode; Indicates the first During the first testing cycle, the measured mechanical structure The resonant frequencies of the first mode; Represents a vector, which is formed by the first... It is composed of all the key mechanical modal parameters extracted during each detection cycle; This represents the baseline mechanical modal parameter vector of health measured and stored when the device is confirmed to be in a healthy state, and its structure is similar to... same; Indicates the first The active mechanical feature vector extracted at the end of each detection cycle has the physical meaning of the current mechanical modal parameter vector. Relative to the health baseline vector The amount of drift.

[0042] This feature vector The technical significance lies in the fact that it quantifies the deviation of physical characteristics caused purely by changes in mechanical structure. Moreover, since it is collected in standby mode, it naturally eliminates the interference of strong electromagnetic fields and control logic delays during the closing operation, providing high-confidence, decoupled evidence for the subsequent Bayesian causal network to distinguish between mechanical and electrical faults.

[0043] The feature extraction module also processes the passive dynamic monitoring data collected by the multimodal sensing module during the device's closing operation to extract passive dynamic features. These passive signals include acoustic timing signals. Vibration timing signal and coil current timing signal .

[0044] The extracted passive dynamic features include: recursive graph quantization features or fractal dimension features extracted from the acoustic timing signal or vibration timing signal based on nonlinear dynamic analysis; and coil response time features or coil driving energy features extracted from the coil current timing signal.

[0045] against and For this module, nonlinear dynamics analysis is specifically introduced to replace traditional time-domain statistics or frequency-domain harmonic analysis. This approach is chosen because nonlinear characteristics are considered more sensitive to subtle dynamic changes in the system's evolution from a healthy state to an early failure.

[0046] The nonlinear dynamic characteristics specifically include recursive graph quantization characteristics and fractal dimension characteristics.

[0047] The method for extracting quantized features from recursive graphs is as follows: First, it is necessary to analyze the time-series signal. Phase space reconstruction is performed to obtain Phase space trajectory of dimension Subsequently, a recursive matrix is ​​constructed. To compare how close each point in the trajectory is to all other points: ; in, and These are any two state points in phase space; It is the time index of the state vector in the phase space trajectory; Indicates time index Phase space state vector at time; Indicates time index Phase space state vector at time; It is a norm; It is a preset distance threshold; It is the Heaviside step function.

[0048] Based on recursive matrix This module calculates its quantitative indicators as quantitative features of the recursive graph, mainly including: Determinism: Characterizes the proportion of diagonal structures in a recursive graph, reflecting the predictability of system dynamics; Entropy: Shannon entropy, which characterizes the distribution of diagonal lengths, reflects the dynamic complexity of a system.

[0049] The method for extracting fractal dimension features is as follows: the Higuchi algorithm can be used to calculate the fractal dimension of a time series signal, which can be used to quantify the self-similarity and irregularity of the signal.

[0050] Acoustic timing signals and vibration timing signal The extracted passive nonlinear features are denoted as ,in This represents the determinism of the acoustic signal. The higher the value, the more predictable the system dynamics reflected by the acoustic signal are, and the lower the randomness. The higher the value, the more regular and predictable the mechanical motion reflected by the vibration signal is; The entropy of the acoustic signal is the higher the value, the more uneven the length distribution of the repeating patterns in the acoustic signal, and the more complex or irregular the dynamic behavior of the system. The entropy of the vibration signal, as above, measures the dynamic complexity reflected by the vibration signal; This is the fractal dimension of the acoustic signal. This value is used to quantify the complexity of the acoustic signal time series. Changes in the friction or impact state between mechanical parts may cause changes in this dimension value. This represents the fractal dimension of the vibration signal, and as above, it is used to quantify the complexity of the vibration signal. These characteristics... reduction or , Changes in these changes can serve as early indicators of wear, loosening, or poor lubrication in mechanical moving parts.

[0051] For high sampling rate coil current timing signals The feature extraction module extracts a set of electrical features. These characteristics are used to characterize the health status of an electromagnetic drive circuit. These characteristics include coil response time characteristics or coil drive energy characteristics.

[0052] Coil response time is defined as the time it takes for the coil current to rise from a low percentage of its peak value to a high percentage. This characteristic primarily reflects the electrical properties of the control circuit, such as capacitor aging and increased resistance.

[0053] The coil drive energy is calculated as the coil current. Throughout the drive action window The integral within the timeframe is used as a proxy for the total work done by the electromagnet: ; in, This represents the driving energy of the coil, which is a proxy indicator of the work done or energy consumed by the electromagnet in one complete driving action. Represents a time variable; Indicates time The timing signal of the changing coil current; This indicates the start time of the integral calculation of the driving energy, i.e., the start time of the selected driving action window; This indicates the end time of the integral calculation of the driving energy, i.e., the end time of the selected driving action window.

[0054] Changes in the driving energy characteristics of a coil can reflect fluctuations in the driving voltage or changes in the coil characteristics.

[0055] The aforementioned passive nonlinear characteristics and passive electrical characteristics together constitute the passive dynamic characteristics.

[0056] Preferably, the feature extraction module will also include environmental parameters collected by the multimodal sensing module, such as temperature. and humidity They are included in the feature set.

[0057] Finally, the feature extraction module in the 1st... After the second closing operation is completed, all the extracted features are assembled to form a complete, multi-dimensional target feature vector at that moment. : ; in, This represents the detection period index, for example, the first... Second test; Represents a vector, which is the vector of the first... The comprehensive state feature vector constructed in each detection cycle and used as input to the subsequent analysis model; This represents the active mechanical feature vector, which, according to the aforementioned definition, represents the amount of drift of the current mechanical modal parameters relative to the healthy baseline; This represents a passive nonlinear eigenvector, which is formed by the first... During a passive dynamic monitoring period, the nonlinear dynamic features extracted from signals such as vibration and acoustics constitute the characteristics. Represents the passive electrical characteristic vector, which is formed by the first... The features extracted from electrical signals such as current and voltage during a passive dynamic monitoring period constitute the composition. Indicates the first The ambient temperature in the environmental reference measured during each testing cycle; Indicates the first The ambient humidity measured in the environmental reference during each detection cycle. This represents the concatenation or chaining operation of vectors or elements, used to combine individual feature vectors and elements into a longer, composite feature vector. ; Represents the structure of a vector or array.

[0058] Target feature vector It will serve as the standard input for subsequent deviation prediction and diagnostic analysis modules, integrating pure mechanical drift, dynamic nonlinearity, and electrical drive characteristics, providing ample data support for achieving high-precision degradation prediction and root cause diagnosis.

[0059] The deviation prediction module uses a symbolic regression algorithm to establish an explicit degradation equation representing the relationship between the target feature vector and the closing time deviation, which inherits the structured target feature vector output by the feature extraction module. And based on the actual measured closing time deviation As a supervisory target, its core task is to establish a mathematical model that can accurately represent the intrinsic relationship between multidimensional features and final performance.

[0060] In this embodiment, the deviation prediction module deliberately avoids conventional, complex models that are prone to falling into the black-box dilemma. This invention not only needs to predict whether a deviation will occur, but also needs to understand how the deviation occurs. Therefore, this module preferably employs the symbolic regression algorithm from genetic programming techniques.

[0061] The technical advantage of using the symbolic regression algorithm is that it does not presuppose the structure of the model, but instead searches in a vast mathematical expression space through an evolutionary computation method. Its aim is to actively discover explicit degradation equations that can fit the observed data, are simple in form, and are physically interpretable.

[0062] The deviation prediction module first establishes an explicit degradation equation characterizing the relationship between the target feature vector and the closing time deviation based on the known target feature vector and the closing time deviation using a symbolic regression algorithm. The specific process is as follows:

[0063] To achieve symbolic regression, we first define the basic elements required for the evolutionary process: Terminal set: This set constitutes the leaf nodes of the candidate mathematical formulas, and its source is the target feature vector output by the feature extraction module. All elements in it; it includes active mechanical characteristics, passive nonlinear characteristics, passive electrical characteristics, etc., and also includes a set of random constants; Function set: This set constitutes the non-leaf nodes of candidate mathematical formulas, i.e., the mathematical operators used to connect the terminals; this set may include Basic mathematical functions are used to ensure that the model has sufficient nonlinear expressive power.

[0064] In this embodiment, the deviation prediction module searches for the optimal formula by simulating the process of biological evolution. First, it randomly generates... An initial population consisting of 10 candidate formulas, where each formula... All by (Terminal set) and The elements in a (function set) are combined and are usually represented as a tree data structure.

[0065] The bias prediction module needs clear evaluation criteria to measure each formula in the population. The fitness function is defined to determine the advantages and disadvantages of fitness. The root mean square error can be used as the fitness function, and its calculation method is as follows: ; in, Indicates the first The candidate formula for the first training samples The predicted value; This is the actual closing time deviation corresponding to this sample; This represents the total number of training samples. The smaller the value, the better the formula. The better the fit to the training data.

[0066] The deviation prediction module then iteratively executes a series of evolutionary operations to guide the population towards a more fitness-optimal direction. These operations mainly include: Selection: Based on fitness To select formulas that perform better with a higher probability. As a parent generation, it is passed down to the next generation; Crossover: Randomly select two parent formulas and swap a subtree in their tree structure to generate a new child formula. This operation is used to combine the best structures in different formulas. Mutation: Randomly changing a node in a formula tree structure, replacing one terminal with another, or replacing one function with another. This operation is used to maintain the diversity of the population and explore new solution spaces.

[0067] After a preset maximum number of iterations Generational evolution, or when the population's optimal fitness When no further significant improvement is observed, the evolutionary process terminates, ultimately producing the individual with the best fitness in the population, denoted as... .Should This is the explicit degradation equation sought by this invention. It clearly reveals, in a concrete and readable mathematical expression, how various key characteristics interact and ultimately lead to the closing time deviation. Things have changed.

[0068] After obtaining the explicit degradation equation, the deviation prediction module also undertakes the task of providing forward-looking condition warnings based on this equation. During subsequent operation of the device, for each new closing operation, the module will receive the latest target feature vector. .

[0069] The module will Substitute into the established explicit degradation equation The predicted closing time deviation is calculated in real time. : ; in, Indicates the first Each detection cycle or time step; Indicates the first The target feature vector collected and constructed in each detection cycle, according to the aforementioned definition, typically contains information such as active mechanical features, passive nonlinear features, passive electrical features, and environmental variables. This represents the optimal explicit degradation equation or optimal prediction model, which is constructed and optimized by the deviation prediction module and is used to map the relationship between the comprehensive state feature vector and the closing time deviation. This indicates that the optimal prediction model is used. The output is in the first position. The predicted closing time deviation value for each detection cycle is an estimate of the time deviation corresponding to the current equipment state by the deviation prediction module, which is used to realize forward-looking state warning.

[0070] Preferably, after obtaining the predicted closing time deviation, the deviation prediction module outputs an abnormal warning based on the predicted closing time deviation, including: Threshold warning: The system pre-sets warning thresholds. and fault threshold Both of these thresholds are within the acceptable range for equipment operation, and If the predicted closing time deviation is greater than the preset warning threshold but less than the preset fault threshold, i.e. This indicates that the device performance has deviated from its optimal health state, and outputs the first abnormal warning to indicate that the phase selection closing device has deviated from its optimal health state. Trend alert: Module cache recent consecutive The predicted closing time deviation sequence of the next step For continuous Linear regression analysis was performed on the predicted closing time deviation for each detection cycle to calculate the slope. If the slope is greater than the preset alarm slope ,Right now This indicates that the closing deviation is showing a continuous and systematic deterioration trend, even though the current predicted value has not yet reached [the expected value]. The system will also output a second abnormal alarm to indicate that the phase selection and closing device is in a deterioration trend.

[0071] The deviation prediction module achieves a shift from passive response to active prediction through the methods described above. Its output... It not only provides operation and maintenance personnel with a transparent and understandable model of equipment state evolution, but its output also... The warning signals also provide crucial input for triggering and decision-making in subsequent diagnostic analysis modules.

[0072] The diagnostic analysis module, as the final execution unit for the accurate location of fault roots in this invention, receives a forward-looking warning signal from the deviation prediction module, or detects an actual closing time deviation. When a deviation from the standard is confirmed, the system can automatically and evidence-based infer the most likely physical cause of the degradation or failure.

[0073] In this embodiment, the diagnostic analysis module uses a Bayesian causal network as its core inference engine. The reason for adopting this technology is that the Bayesian causal network can organically integrate the complex physical mechanism of the phase selection and closing device, the causal relationships between components, and data-driven probabilistic statistics.

[0074] The primary task of the diagnostic analysis module is to construct a directed acyclic graph (DAG) that reflects the fault mechanism of the equipment, denoted as . ,in For a set of nodes, It is an edge set.

[0075] Preferably, the nodes of the Bayesian causal network utilized by the diagnostic analysis module include: The root node representing the fault hypothesis These nodes are the targets of the diagnosis; they have no parent node and represent the original cause of the fault; the root node includes mechanical faults. Electrical faults and sensor failure ; Intermediate nodes representing characteristic evidence These nodes are the basis for diagnosis, and their state is observable. They are discretized representations of the feature vectors output by the feature extraction module; the intermediate nodes include the active mechanical features and the passive dynamic features, i.e., the aforementioned... , as well as ; The leaf node representing the closing time deviation represents the final observed equipment phenomenon, i.e., the closing deviation exceeding the limit event. .

[0076] The causal relationships in the Bayesian causal network are defined as follows: There is a strong causal relationship between the root node of the mechanical fault and the intermediate node of the active mechanical feature; There is no causal relationship between the electrical fault root node and the active mechanical feature intermediate node.

[0077] Specifically, edge sets representing causal relationships The definition of these directed edges is key to achieving accurate diagnosis in this invention. These directed edges represent the physical causal relationships between nodes, and their construction strictly follows the technical solution of this invention, particularly utilizing the information decoupling brought about by active detection. It defines mechanical failure as the direct cause of abnormalities in active mechanical characteristics; It is defined that mechanical failure is also a cause of abnormal passive nonlinear characteristics; It defines electrical faults as the direct cause of abnormalities in passive electrical characteristics; This is a crucial and mandatory definition of non-causality; its physical meaning is that electrical faults will not cause changes in the active mechanical characteristics measured in standby mode. This causal decoupling based on physical mechanisms is the core of this module's ability to distinguish between mechanical and electrical faults. : Defines anomalies in characteristic evidence ( ) or a malfunction in the sensor itself ( This is what causes the final closing deviation to exceed the standard. The common reason for this.

[0078] In network topology After the definition is complete, the diagnostic analysis module needs to parameterize the network, that is, for each node. Define its conditional probability table. This conditional probability table quantitatively describes the probability distribution given its parent node. In this state, the node The probability of taking different states is denoted as . .

[0079] The parameters of the conditional probability table can be learned through statistical study of historical fault data; in cases where data is scarce, they can also be set in conjunction with expert knowledge. It will be assigned a high probability value, and It will be assigned an extremely low probability value in order to solidify. This strong a priori knowledge.

[0080] When the diagnostic analysis module is triggered, it will execute the causal inference step. First, the module will take the latest feature vector output by the feature extraction module. Discretization yields a set of observed evidence. .

[0081] The diagnostic analysis module randomly selects this evidence. Instantiated into a Bayesian causal network, and using the Bayesian inversion theorem, computed the current evidence. Under the conditions that have already occurred, the root causes of each failure The posterior probability of occurrence. The mathematical principle behind this inference is as follows: ; in, In the absence of any evidence, various malfunctions were identified. The prior probability; Under specific failure assumptions, the current evidence was observed. The likelihood probability; This is the updated evidence required by this module. The subsequent posterior probability. This calculation can be performed efficiently using inference methods such as joint tree algorithms.

[0082] The final output of the diagnostic analysis module is generated based on the maximum a posteriori probability criterion. The module compares the posterior probabilities of all root node failures and selects the one with the highest probability value as the most probable root cause of the event. : ; in, This indicates the final output of the fault root cause diagnosis conclusion; Represents the i-th possible device state or fault root cause mode. It represents a discrete set that contains all predefined device states or fault root cause patterns that need to be diagnosed; It is a mathematical operator that represents all elements belonging to the set. elements In the middle, find the expression that makes the subsequent probability expression The element that gets the maximum value ; This represents evidence, namely all relevant feature information extracted from sensor data upon which the diagnostic analysis module relies; Represents the posterior probability, whose physical meaning is: in the event of observing specific evidence... Under these conditions, the true state of the equipment or the root cause of the fault is The diagnostic model calculates the posterior probability for each possible state.

[0083] If the inference result is If the system fails to detect a mechanical fault, it will output a mechanical fault as a diagnostic conclusion. This conclusion is highly reliable because of the observed evidence. It is very likely that it contains abnormal active mechanical characteristics but normal passive electrical characteristics, which is consistent with... The causal framework is highly consistent. In this way, this module enables deep and interpretable diagnosis, from the device's symptoms to the underlying nature of the fault.

[0084] Reference Figure 2 The present invention also provides an automatic detection method for a phase selection and closing device based on an intelligent algorithm, comprising: S1: When the phase selection closing device is in standby mode, apply an active excitation signal and acquire an active mechanical response signal; and acquire a passive operation signal when the phase selection closing device is in closing operation. S2: Extract active mechanical features from the active excitation signal and the active mechanical response signal, extract passive dynamic features from the passive operation signal, and concatenate the active mechanical features and the passive dynamic features into a target feature vector; S3: After establishing an explicit degradation equation representing the relationship between the target feature vector and the closing time deviation using the symbolic regression algorithm, the target feature vector is input into the explicit degradation equation to obtain the predicted closing time deviation, and an abnormal warning is output based on the predicted closing time deviation. S4: When the deviation prediction module outputs an abnormal warning, the root cause of the fault in the phase selection and closing device is inferred based on the target feature vector and using a Bayesian causal network.

[0085] In summary, the automatic detection system for phase selection and closing device based on intelligent algorithm described in this invention has the following beneficial effects:

[0086] 1. This invention achieves effective separation of the purely mechanical characteristics and electromechanical coupling dynamic characteristics of the equipment by using active harmonic detection when the device is in standby mode and passive dynamic monitoring when the device is closed. This active-passive combined data acquisition strategy overcomes the problem of confusion between mechanical signals and electromagnetic interference signals in traditional methods, and provides a high-confidence, decoupled data foundation for subsequent accurate fault root cause diagnosis.

[0087] 2. This invention introduces nonlinear dynamic analysis to process passively monitored vibration and acoustic signals, which can capture the subtle dynamic changes in the early stage of the device's evolution from health to failure. Compared with traditional time-frequency domain statistical features, these nonlinear features are more sensitive to the reduction of motion determinism or changes in system complexity, thereby improving the system's ability to perceive early potential failures and achieving earlier forward-looking warnings.

[0088] 3. This invention utilizes explicit degradation equations for real-time deviation prediction and sets up a dual-mode early warning mechanism that combines trend warning and threshold warning, thereby achieving forward-looking management of equipment status. The system can identify the continuous degradation trend or slight deviation of performance in advance before the actual closing time deviation exceeds the qualified range, thus transforming traditional post-fault maintenance into state-based predictive maintenance, effectively avoiding unexpected downtime and ensuring the operational reliability of the phase selection closing device.

[0089] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0091] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0093] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. An automatic detection system for a phase selection and closing device based on intelligent algorithms, characterized in that, include: The multimodal sensing module is used to apply an active excitation signal and acquire an active mechanical response signal when the phase selection closing device is in standby mode. And acquire the passive operation signal when the phase selection closing device is closed; The feature extraction module, connected to the multimodal perception module, is used to extract active mechanical features from the active excitation signal and the active mechanical response signal, extract passive dynamic features from the passive operation signal, and concatenate the active mechanical features and the passive dynamic features into a target feature vector. The deviation prediction module, connected to the feature extraction module, is used to establish an explicit degradation equation representing the relationship between the target feature vector and the closing time deviation using a symbolic regression algorithm, input the target feature vector into the explicit degradation equation to obtain the predicted closing time deviation, and output an abnormal warning based on the predicted closing time deviation. The diagnostic analysis module, connected to the feature extraction module and the deviation prediction module, is used to infer the root cause of the fault in the phase selection and closing device based on the target feature vector and using a Bayesian causal network when the deviation prediction module outputs an abnormal warning.

2. The automatic detection system for a phase selection and closing device based on an intelligent algorithm according to claim 1, characterized in that, The multimodal sensing module includes: The active excitation unit, including a piezoelectric ceramic exciter, is used to apply an active excitation signal to the mechanical structure when the phase selection closing device is in standby mode. The response pickup unit includes a wideband vibration sensor, which is used to acquire the active mechanical response signal generated by the active excitation signal of the mechanical structure when the phase selection closing device is in standby mode.

3. The automatic detection system for a phase selection and closing device based on an intelligent algorithm according to claim 1, characterized in that, The multimodal sensing module also includes: The environmental sensing unit is used to acquire environmental parameters of the environment in which the phase selection and closing device is located.

4. The automatic detection system for a phase selection and closing device based on an intelligent algorithm according to claim 1, characterized in that, The method for extracting active mechanical features from the active excitation signal and the active mechanical response signal is as follows: Fourier transforms are performed on the active excitation signal and the active mechanical response signal respectively, and the ratio of the transformed active mechanical response signal to the transformed active excitation signal is used as the mechanical transfer function of the phase selection and closing device in the frequency domain. Based on the mechanical transfer function of the baseline state, obtain the baseline mechanical modal parameters; Based on the mechanical transfer function of the current detection cycle, obtain the mechanical modal parameters of the current detection cycle; The difference between the mechanical modal parameters of the current detection cycle and the corresponding baseline mechanical modal parameters is used as the active mechanical feature.

5. The automatic detection system for a phase selection and closing device based on an intelligent algorithm according to claim 1, characterized in that, The passive operation signals include acoustic timing signals, vibration timing signals, and coil current timing signals.

6. The automatic detection system for a phase selection and closing device based on an intelligent algorithm according to claim 5, characterized in that, The passive dynamic features extracted by the feature extraction module include: Based on nonlinear dynamics analysis, recursive graph quantization features or fractal dimension features are extracted from the acoustic time-series signal or vibration time-series signal.

7. The automatic detection system for a phase selection and closing device based on an intelligent algorithm according to claim 5, characterized in that, The passive dynamic features extracted by the feature extraction module also include: The coil response time characteristics or coil drive energy characteristics extracted from the coil current timing signal.

8. The automatic detection system for a phase selection and closing device based on an intelligent algorithm according to claim 1, characterized in that, The deviation prediction module outputs an anomaly warning based on the predicted closing time deviation, including: If the predicted closing time deviation is greater than the preset warning threshold but less than the preset fault threshold, the first abnormal warning is output to indicate that the phase selection closing device has deviated from the optimal health state. Linear regression analysis is performed on the predicted closing time deviation of multiple consecutive detection cycles to calculate the slope; if the slope is greater than the preset alarm slope, a second abnormal alarm is output to indicate that the phase selection closing device is in a deterioration trend.

9. The automatic detection system for a phase selection and closing device based on an intelligent algorithm according to claim 1, characterized in that, The nodes of the Bayesian causal network used in the diagnostic analysis module include: The root node represents the fault hypothesis, including mechanical faults, electrical faults, and sensor faults; The intermediate node representing characteristic evidence includes the active mechanical feature and the passive dynamic feature; Leaf node representing closing time deviation.

10. An automatic detection method for a phase-selective closing device based on an intelligent algorithm, used in the automatic detection system for a phase-selective closing device based on an intelligent algorithm as described in any one of claims 1-9, characterized in that, Includes the following steps: When the phase selection closing device is in standby mode, an active excitation signal is applied and an active mechanical response signal is acquired; And acquire the passive operation signal when the phase selection closing device is closed; Active mechanical features are extracted from the active excitation signal and the active mechanical response signal, and passive dynamic features are extracted from the passive operation signal. The active mechanical features and the passive dynamic features are then concatenated into a target feature vector. After establishing an explicit degradation equation representing the relationship between the target feature vector and the closing time deviation using the symbolic regression algorithm, the target feature vector is input into the explicit degradation equation to obtain the predicted closing time deviation, and an abnormal warning is output based on the predicted closing time deviation. When the deviation prediction module outputs an abnormal warning, the root cause of the fault in the phase selection and closing device is inferred based on the target feature vector and using a Bayesian causal network.

Images