A pre-diagnosis based device failure prediction system

Abstract

This invention relates to the field of electrical equipment fault prediction and intelligent diagnosis technology, specifically a pre-diagnosis-based equipment fault prediction system, comprising: acquiring high-frequency current or voltage time series of the main circuit and calculating real-time variable load current gradient sequence; determining the length of structural elements based on the gradient, obtaining the load baseline using a morphological combination filter and differentially obtaining a minor fault signal; reconstructing the phase space of the minor fault signal, and assessing the fault risk value by combining the reference parameter matrix, characteristic offset, and discrete change rate; generating a fault warning command or circuit breaker control command based on the risk value change within the monitoring window, and outputting it to the alarm terminal and circuit breaker execution device; this invention can pre-diagnose early insulation degradation, minor discharge, or weak leakage current changes.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment fault prediction and intelligent diagnosis technology, specifically to a equipment fault prediction system based on pre-diagnosis. Background Technology

[0002] Existing electrical equipment fault prediction systems typically sample main circuit current, voltage, or insulation parameters and judge the equipment's operating status based on fixed thresholds, single time-domain features, or conventional trend quantities. In related technologies, to identify whether the equipment has insulation degradation, micro-discharge, or leakage abnormalities, the amplitude of the collected high-frequency signals can usually be directly compared, or the signals can be filtered first before extracting features and performing status assessment. However, when electrical equipment is frequently started and stopped, the load changes continuously, or a nonlinear load is connected, the high-frequency signal of the main circuit often contains waveform disturbances caused by load changes and weak abnormal components corresponding to early faults. The former method is prone to misjudging load fluctuations as faults, while the latter method, when the filtering scale is fixed or the feature analysis parameters lack adaptive constraints, is difficult to balance the accuracy of weak fault extraction with the stability of early warning. Therefore, equipment failure prediction methods in related technologies are difficult to simultaneously achieve effective isolation, stable assessment, and reliable early warning of early minor faults under complex variable load conditions. Summary of the Invention

[0003] The purpose of this invention is to provide a pre-diagnosis-based equipment fault prediction system that avoids confusing waveform abrupt changes caused by load variations with actual fault characteristics when equipment is frequently started and stopped or when there are nonlinear load variations. It can also dynamically adjust the filtering scale to accurately separate the load baseline and extract minute fault signals by adaptively adjusting the load level. Furthermore, it combines multi-dimensional state point sequences for comprehensive evaluation within the monitoring window to achieve more stable and accurate early fault warning and circuit breaker control.

[0004] The objective of this invention can be achieved through the following technical solutions: A pre-diagnosis-based equipment failure prediction system includes: The data acquisition module is used to acquire the high-frequency current or voltage time series of the main circuit of the electrical equipment and simultaneously calculate the real-time variable load current gradient sequence of the main circuit. The morphological filtering module is used to determine the length of the structural element based on the real-time variable load current gradient sequence, and to filter the high-frequency current or voltage time series using a morphological combination filter based on the structural element length to obtain the load baseline. Then, it performs a differential operation with the high-frequency current or voltage time series to obtain the minor fault signal. The state feature analysis module is used to establish a reference parameter matrix based on the high-frequency current or voltage time series during normal operation; determine the dimension and delay time of phase space reconstruction, and reconstruct the phase space of the minor fault signal according to the dimension and delay time, mapping it into a multi-dimensional state point sequence; calculate the difference between the current parameter matrix and the reference parameter matrix of the sequence under the current monitoring window to obtain the feature offset; and track the distance change of multiple state point pairs in the sequence whose initial distance is not greater than the preset neighborhood radius after a preset time step to obtain the discrete rate of change. The fault assessment module is used to determine the fault risk value based on the feature offset and discrete change rate, and generate fault warning instructions or circuit breaker control instructions according to their changes within the monitoring window, as well as trigger locking or adjustment operations on at least one of the following processing parameters: structural element length, dimension, delay time, and preset neighborhood radius. The control output module is used to send fault warning commands and circuit breaker control commands to the alarm terminal and the circuit breaker execution device, respectively.

[0005] Preferably, the data acquisition module is specifically used for: Collect the power frequency main circuit current sampling value of the main circuit; Extract the variation characteristics of the main circuit current sample values ​​at adjacent sampling times; based on the correlation between the variation characteristics and the sampling time interval, determine the real-time variable load current gradient sequence at the corresponding sampling time.

[0006] Preferably, when determining the length of the structural element based on the real-time variable load current gradient sequence, the morphological filtering module is specifically used for: Obtain the basic structural element length, rated maximum current gradient threshold, filter attenuation coefficient, and minimum structural element length lower limit of the electrical equipment; The attenuation factor is determined based on the mapping relationship between the real-time variable load current gradient sequence, the rated maximum current gradient threshold, and the filter attenuation coefficient; the length of the candidate structural element is determined based on the length of the basic structural element and the attenuation factor. The maximum value between the candidate structuring element length and the minimum structuring element length lower limit is selected as the structuring element length.

[0007] Preferably, when the morphological filtering module uses a morphological combined filter with a structuring element length to filter a high-frequency current or voltage time series to obtain a load baseline, it is specifically used for: A flat structural element is constructed based on the structural element length; morphological opening and closing operations are sequentially performed on the high-frequency current or voltage time series using the flat structural element to obtain the opening and closing filter sequence; By using flat structural elements constructed based on the length of the structural elements, morphological closing and opening operations are sequentially performed on high-frequency current or voltage time series to obtain a closed-open filter sequence. The load baseline is obtained by adding the open-closed filter sequence to the closed-open filter sequence and dividing by two.

[0008] Preferably, when the state feature analysis module reconstructs the phase space of the minor fault signal and maps it into a multi-dimensional state point sequence, it is specifically used for: The pre-selected dimension is determined by the nearest neighbor method, and the pre-selection delay time is determined by correlation analysis. Obtain the length of the structuring element output by the morphological filtering module in the current monitoring window, and calculate the average structuring element length; By combining the average structural element length and constraining the pre-selected dimensions and pre-selected delay times according to a preset proportional relationship, the dimensions and delay times are determined. Based on the dimension and delay time, a one-dimensional micro-fault signal is reconstructed into a multi-dimensional state point sequence.

[0009] Preferably, when the state feature analysis module calculates the difference between the current parameter matrix and the reference parameter matrix of the multidimensional state point sequence under the current monitoring window to obtain the feature offset, it is specifically used for: Feature extraction is performed on the baseline parameter matrix and the current parameter matrix under the current monitoring window to obtain the baseline feature set and the current feature set and their corresponding central quantities; based on the degree of numerical deviation between the current feature set and the baseline feature set, and the spatial offset relationship between the corresponding central quantities, feature fusion calculation is performed to obtain the feature offset.

[0010] Preferably, when the state feature analysis module tracks the distance change between two state points in a multidimensional state point sequence whose initial distance is no greater than a preset neighborhood radius after a preset time step, and obtains the discrete rate of change, it is specifically used for: In a multidimensional sequence of state points, select multiple pairs of state points whose initial distance is no greater than the neighborhood radius; Local change features are extracted based on the ratio of the distance between the evolved state points and the initial distance after a time step; the discrete change rate is obtained by combining the local change features of multiple state point pairs.

[0011] Preferably, when determining the fault risk value based on the feature offset and the discrete rate of change, the fault assessment module is specifically used for: Read the preset feature offset threshold and discrete change rate threshold from the threshold library corresponding to the device model; The fault risk value is obtained by weighting the deviations of the feature offset from the feature offset threshold and the deviations of the discrete change rate from the discrete change rate threshold.

[0012] Preferably, when the fault assessment module generates a fault warning command or circuit breaker control command based on the changes in the fault risk value within the monitoring window, and triggers a locking or adjustment operation on at least one of the processing parameters among the length, dimension, delay time, and neighborhood radius of the structural element based on the changes, it is specifically used for: The fault risk value corresponding to each monitoring window is calculated using a sliding monitoring window method, and a risk value sequence is constructed. Configure preset normal threshold, early warning threshold and fault deterioration warning threshold, and the normal threshold is less than the early warning threshold, and the early warning threshold is less than the fault deterioration warning threshold. When continuous When the fault risk value within a monitoring window is greater than the early warning threshold but less than or equal to the fault deterioration warning threshold, the electrical equipment is determined to be in an early fault trend, and a fault warning command is generated. It is a natural number greater than 1; the length, dimension, delay time and neighborhood radius of the structure element corresponding to the current monitoring window are locked, and a suspected fault parameter group is formed for archiving; the difference between the feature offset of the current monitoring window and the previous monitoring window is calculated to determine the feature change direction, and the feature offset, discrete change rate, feature change direction and corresponding timestamp of the current monitoring window are written into the pre-established historical record library; When continuous When the fault risk value within a monitoring window is greater than the normal threshold but less than or equal to the early warning threshold, the electrical equipment is determined to be in observation status, and the database status flag is updated. When continuous When the fault risk value within a monitoring window is less than or equal to the normal threshold, the electrical equipment is determined to be in normal condition, and the database status flag is updated.

[0013] Preferably, the fault assessment module is also used for: When continuous When the fault risk value in a monitoring window exceeds the fault deterioration warning threshold, the fault is determined to have entered the continuous deterioration stage, and a circuit breaker control command is generated. Extract the most recent normal operation window data and the current abnormal window data before this failure, and write them into a pre-established fault record library in pairs. This allows the fault assessment module to update the feature offset threshold, discrete rate of change threshold, and corresponding feature offset weight and rate of change weight based on the paired data. After the electrical equipment resumes operation, when continuous When the fault risk value within a monitoring window is less than or equal to the normal threshold, the window data under normal conditions is extracted, and the baseline parameter matrix is ​​updated in real time.

[0014] The beneficial effects of this invention are: 1. This invention synchronously calculates the real-time variable load current gradient sequence of the main circuit and dynamically determines the length of the structural element accordingly. Then, it uses a morphological combination filter to extract the load baseline and minor fault signals. This design effectively overcomes the defect of the existing technology that the fixed filter scale is prone to misjudging the waveform disturbance caused by load changes as a fault when there are frequent starts and stops or continuous load changes, and ensures the accuracy of early separation and extraction of weak abnormal components. 2. This invention reconstructs the phase space of minute fault signals according to dimension and delay time, mapping them into a multi-dimensional state point sequence, and calculates the feature offset and discrete rate of change in combination with the reference parameter matrix during normal operation. This scheme changes the limitation of the traditional single time-domain feature definition of early weak faults being not clear enough. It extracts features by tracking distance evolution in multi-dimensional space and comparing matrix differences, which significantly improves the stability of early fault feature assessment of equipment. 3. This invention determines the fault risk value based on the feature offset and discrete change rate, generates an early warning or circuit breaker command based on its changes within the monitoring window, and triggers the locking operation of processing parameters such as the length and dimension of structural elements when a fault is suspected. This not only avoids the risk of misjudgment caused by a single disturbance, but also ensures the consistency of the feature analysis standard during the continuous evolution of the fault, thereby achieving more reliable early warning and circuit breaker protection under complex working conditions. Attached Figure Description

[0015] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A block diagram of a device failure prediction system based on pre-diagnosis, according to an embodiment of the present disclosure, is shown. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 A pre-diagnosis-based equipment fault prediction system includes: a data acquisition module for acquiring high-frequency current or voltage time series of the main circuit of electrical equipment and simultaneously calculating the real-time variable load current gradient sequence of the main circuit; The morphological filtering module is used to determine the length of the structural element based on the real-time variable load current gradient sequence, and to filter the high-frequency current or voltage time series using a morphological combination filter based on the structural element length to obtain the load baseline. Then, it performs a differential operation with the high-frequency current or voltage time series to obtain the minor fault signal. The state feature analysis module is used to establish a reference parameter matrix based on the high-frequency current or voltage time series during normal operation; determine the dimension and delay time of phase space reconstruction, and reconstruct the phase space of the minor fault signal according to the dimension and delay time, mapping it into a multi-dimensional state point sequence; calculate the difference between the current parameter matrix and the reference parameter matrix of the sequence under the current monitoring window to obtain the feature offset; and track the distance change of multiple state point pairs in the sequence whose initial distance is not greater than the preset neighborhood radius after a preset time step to obtain the discrete rate of change. The fault assessment module is used to determine the fault risk value based on the feature offset and discrete change rate, and generate fault warning instructions or circuit breaker control instructions according to their changes within the monitoring window, as well as trigger locking or adjustment operations on at least one of the following processing parameters: structural element length, dimension, delay time, and preset neighborhood radius. The control output module is used to send fault warning commands and circuit breaker control commands to the alarm terminal and the circuit breaker execution device, respectively. The data acquisition module is specifically used for: acquiring the power frequency main circuit current sampling value of the main circuit; extracting the change characteristics of the power frequency main circuit current sampling value at adjacent sampling times; and determining the real-time variable load current gradient sequence at the corresponding sampling time based on the correlation between the change characteristics and the sampling time interval. Divide the absolute value by twice the sampling time interval to obtain the real-time variable load current gradient sequence corresponding to the current sampling moment; When determining the length of structural elements based on the real-time variable load current gradient sequence, the morphological filtering module is specifically used to: obtain the basic structural element length of electrical equipment, the rated maximum current gradient threshold, the filtering attenuation coefficient, and the minimum structural element length lower limit. The attenuation factor is determined based on the mapping relationship between the real-time variable load current gradient sequence, the rated maximum current gradient threshold, and the filter attenuation coefficient; the length of the candidate structural element is determined based on the length of the basic structural element and the attenuation factor; the maximum value between the length of the candidate structural element and the minimum lower limit of the structural element length is selected as the length of the structural element.

[0018] To further explain: This system is deployed in the main circuit of an industrial variable frequency motor that is frequently started and stopped and connected to a nonlinear load, and is used to pre-diagnose early insulation degradation, minor discharge, or weak leakage current changes in the main circuit; The load on the main circuit is in a state of continuous change. The high-frequency sampling signal contains both the variable carrier waveform caused by the operation of the equipment itself and the weak abnormal components that may correspond to the fault. If the threshold is directly applied to the original high-frequency sequence, it is easy to confuse the sudden change caused by the load change with the fault characteristics. This system has constructed a processing link that connects data acquisition, load variation characterization, baseline separation, state feature analysis, risk assessment to control output in sequence. During system operation, the data acquisition module adopts a dual-rate clock synchronization mechanism to acquire the high-frequency current time series or high-frequency voltage time series of the main circuit, and at the same time reads the power frequency main circuit current sampling value of adjacent power frequency sampling times before and after the current power frequency sampling time from the power frequency sampling branch of the main circuit. Since the time granularity of high-frequency sampling is smaller than that of the power frequency cycle, if the power frequency branch is directly required to update synchronously at a high frequency rate, it will cause the physical hardware to be unable to respond and the gradient calculation to have a time scale deviation that exceeds the hardware tolerance limit. Therefore, the real-time variable load current gradient corresponding to the current power frequency sampling moment is actually a low-frequency reference value calculated by dividing the absolute value of the difference between the current power frequency main circuit current sampling value and the sampling value at the previous adjacent power frequency sampling moment by the power frequency sampling time interval. The system performs zero-order hold mapping to synchronously transfer the low-frequency reference value to all high-frequency sampling points within the corresponding power frequency monitoring window, thereby expanding to obtain the real-time variable load current gradient sequence corresponding to the current high-frequency sampling moment. The resulting gradient sequence and the high-frequency time series maintain a point-to-point correspondence within the same monitoring process and are written together into the data cache of the current monitoring window for subsequent filtering modules to use. If the power frequency sampling value is not completely obtained within a certain power frequency sampling period, the current period will not directly generate a new gradient value, but will retain the gradient record formed in the previous effective period and mark the period as missing, so that the morphological filtering module can continue to execute and the entire monitoring link will not be interrupted due to a single sampling gap. When filtering is performed solely based on the length of fixed structural elements, if a sudden load is applied to the main circuit of the motor or the operating state is frequently switched, the filtering scale will not be consistent with the degree of load variation, which may cause the load waveform envelope to enter the subsequent fault analysis process. To avoid this situation, the morphological filtering module first reads the length of the basic structural element, the rated maximum current gradient threshold, the filtering attenuation coefficient, and the minimum structural element length lower limit of the device. The ratio of the real-time variable load current gradient sequence to the rated maximum current gradient threshold is used as the current operating condition quantity. The negative exponent of the product of this ratio and the filter attenuation coefficient is calculated to form the attenuation factor. The length of the basic structural element is then multiplied by the attenuation factor and rounded down to obtain the length of the candidate structural element. If the length of the candidate structural element is less than the minimum structural element length lower limit, then the minimum structural element length lower limit is maintained as the structural element length for this cycle. After this processing, the structural element length is shortened when the load increases and rises when the operating condition stabilizes, so that the same set of morphological filters can adapt to different load stages. After determining the length of the structural element, the morphological filtering module uses a morphological combination filter of that length to process high-frequency current or voltage time series and form a load baseline. The original high-frequency time series and the load baseline are differentially processed within the same monitoring window to obtain a minor fault signal. This minor fault signal is no longer written back to the original sampling buffer, but is stored separately in the feature analysis buffer, which is associated with the length of the structure element and the window identifier corresponding to this window, so that the state feature analysis module can directly call it. If the length of the high-frequency sequence in a certain window is insufficient due to sampling jitter and cannot meet the analysis requirements of the current window, then the window will only retain the original record, will not perform phase space analysis, and will wait for the next complete window to continue processing; The status feature analysis module pre-establishes a reference parameter matrix based on the high-frequency current or voltage time series during normal equipment operation; this reference parameter matrix is ​​stored in the reference model storage area as a health reference and is not overwritten by the original sampling in subsequent monitoring windows. For the current window, the module reads the aforementioned minor fault signal, reconstructs the phase space according to the dimension and delay time, and maps the one-dimensional signal into a multi-dimensional state point sequence; The specific reconstruction process is as follows: starting with the current sampling point in the micro-fault signal sequence, the subsequent sampling values ​​are extracted sequentially according to the delay time interval until the number of extractions reaches the dimension, thus forming a multi-dimensional state point; as the starting point slides point by point, multiple multi-dimensional state points are generated in sequence, forming a multi-dimensional state point sequence. Based on this state point sequence, after horizontally arranging and combining each multidimensional state point to form the current parameter matrix, the difference between the current parameter matrix and the reference parameter matrix is ​​further calculated to obtain the feature offset. At the same time, the module selects state point pairs whose initial distance is no greater than the preset neighborhood radius from the same set of multidimensional state point sequences, and tracks their distance changes after a preset time step to obtain discrete change rate; Thus, the current window forms two feature quantities for fault assessment: feature offset and discrete rate of change. If the number of state point pairs that meet the neighborhood radius condition in a certain window is insufficient, the discrete rate of change of that window will not be used alone for trigger control, but will participate in subsequent trend judgment together with the result of the previous effective window, so as to avoid misjudgment of a single window due to the lack of occasional point pairs. The fault assessment module receives the feature offset and discrete change rate of the current window from the status feature analysis module, calculates the fault risk value, maintains the risk value sequence according to the monitoring window, and makes a status determination based on the change of the risk value within the monitoring window. When the risk value reaches the preset warning condition within the window, the module generates a fault warning command; when the risk value reaches the circuit breaker condition within the window, the module generates a circuit breaker control command. At the same time, the fault assessment module also performs a locking or adjustment operation on at least one of the processing parameters related to the current judgment: length of the structural element, set dimension, set delay time, and neighborhood radius. The results after parameter locking are written to the suspected fault parameter record area to ensure that subsequent windows maintain consistent judgment criteria at the same abnormal evolution stage; the results after parameter adjustment are updated to the running parameter area for calculation and calling in subsequent new windows. The control output module receives the control results issued by the fault assessment module, sends the fault warning command to the pre-connected alarm terminal, and sends the circuit breaker control command to the pre-connected circuit breaker execution device. Under this industrial variable frequency motor, the alarm terminal is used to prompt the operation and maintenance side to pay attention to the current equipment status, and the circuit breaker is used to cut off the main circuit under the condition that the fault continues to worsen; if the alarm terminal communication is abnormal, the warning command is at least retained in the local event log. If the circuit breaker does not return an execution confirmation, the system maintains the current fault status record and continues to output the risk assessment results for subsequent cycles for review by the upper control link.

[0019] In a preferred embodiment of the present invention, when the morphological filtering module uses a morphological combined filter with a structuring element length to filter a high-frequency current or voltage time series to obtain a load baseline, it is specifically used for: A flat structural element is constructed based on the structural element length; morphological opening and closing operations are sequentially performed on the high-frequency current or voltage time series using the flat structural element to obtain the opening and closing filter sequence; Using a flat structural element constructed based on the structural element length, morphological closing and opening operations are sequentially performed on the high-frequency current or voltage time series to obtain a closed-open filter sequence; the open-closed filter sequence is added to the closed-open filter sequence and divided by two to obtain the load baseline; When reconstructing the phase space of minute fault signals and mapping them into a multi-dimensional state point sequence, the state feature analysis module is specifically used for: determining the pre-selected dimension through the nearest neighbor method and determining the pre-selected delay time through correlation analysis; obtaining the length of the structural element output by the morphological filtering module under the current monitoring window and calculating the average structural element length. By combining the average structural element length and constraining the pre-selected dimension and pre-selected delay time according to the preset ratio, the dimension and delay time are determined; according to the dimension and delay time, the one-dimensional micro-fault signal is reconstructed into a multi-dimensional state point sequence. When calculating the difference between the current parameter matrix and the reference parameter matrix of the multidimensional state point sequence under the current monitoring window to obtain the feature offset, the state feature analysis module is specifically used to: extract features from the reference parameter matrix and the current parameter matrix under the current monitoring window respectively to obtain the reference feature set and the current feature set and their corresponding central quantities; and perform feature fusion calculation to obtain the feature offset based on the degree of numerical deviation between the current feature set and the reference feature set, as well as the spatial offset relationship between the corresponding central quantities. When the state feature analysis module tracks the distance change of two state points whose initial distance is no greater than the preset neighborhood radius in a multidimensional state point sequence and obtains the discrete rate of change, it is specifically used to: filter out multiple pairs of state points whose initial distance is no greater than the neighborhood radius in a multidimensional state point sequence. Local change features are extracted based on the ratio of the distance between the evolved state points and the initial distance after a time step; the discrete change rate is obtained by combining the local change features of multiple state point pairs.

[0020] For the main circuit of a variable frequency motor that frequently starts and stops and is connected to a nonlinear load, the process of separating minor fault signals and converting state characteristics includes: constructing a load baseline in a high-frequency sequence where the load waveform and abnormal components overlap, and converting minor fault signals into judgment feature quantities based on multi-dimensional state point sequences and their discrete change relationships. When using only the morphological filtering result of a single path as the baseline, it is easily affected by the bias of one-sided operation. Especially when the motor load change rate exceeds the preset load fluctuation threshold and there are pulse disturbances in the high-frequency sequence, the baseline will have the problem of unbalanced tracking of local peaks and valleys. To mitigate this impact, the morphological filtering module constructs flattened structural elements within the current window based on the determined structural element length, and executes two processing paths for the high-frequency current or voltage time series respectively. In the first path, morphological opening is performed first, followed by morphological closing, to obtain the opening and closing filter sequence; In the second path, morphological closing operation is performed first, followed by morphological opening operation, to obtain the closing-opening filter sequence; the average of the two sets of sequences is taken to form the load baseline of the current window; The load baseline is stored in a one-to-one correspondence with the original high-frequency time series, and a small fault signal is formed by difference. Since the load baseline has been uniformly calculated within the window, subsequent phase space processing no longer backtracks the original morphological operation process, but only reads the difference result and the length record of the structural element of the window, thereby maintaining the separation between the front and back stage processing. If the phase space reconstruction parameters are arbitrarily set directly for the obtained small fault signals, the reconstruction scale may be inconsistent with the previous filtering scale, resulting in unstable characterization of the same fault pulse in different windows. Therefore, the state feature analysis module first determines the pre-selected delay time through correlation analysis, and then determines the pre-selected dimension through the nearest neighbor method; In practical implementation, the system calculates the autocorrelation function of the one-dimensional micro-fault signal under different delay step sizes. The discrete calculation formula for the autocorrelation function is as follows: , in, To delay the step size, The length of the signal sequence within the current monitoring window. For the first The amplitude of a tiny fault signal at each sampling point The arithmetic mean of the signal sequence; extract The delay step corresponding to the first drop of the calculated value to the preset initial attenuation ratio is used as the pre-selected delay time to ensure that each coordinate component of the reconstructed phase space has sufficient independence. The system calculates the distance between adjacent state points under different reconstruction dimensions. If an increase in dimension causes the relative increase in the distance between adjacent points to exceed a preset false nearest neighbor threshold, then the point pair is marked as a false nearest neighbor. The system calculates the overall proportion of false nearest neighbors and selects those that reduce this proportion to, for example, a lower tolerance limit. The minimum dimension at the following time is used as the pre-selected dimension; At the same time, the length of the structural element corresponding to each sampling segment in the current monitoring window is read from the morphological filtering module, and the average structural element length is calculated. The average structuring element length reflects the overall scale used for load baseline separation within the current window; the system then combines the preset proportional relationship to constrain and filter the pre-selected dimensions and pre-selected delay times, forming the set dimensions and set delay times actually used for the calculation of the current window; The specific constraint filtering logic is as follows: The system is pre-configured with dimension scaling factor and delay scaling factor. The average structuring element length is multiplied by the dimension scaling factor and delay scaling factor respectively to obtain the upper limit of dimension constraint and the upper limit of delay constraint. Determine if the pre-selected dimension is greater than the upper limit of the dimension constraint. If it is, force the dimension to be the upper limit of the dimension constraint and round it down. If it is not greater, directly use the pre-selected dimension as the dimension. Similarly, determine whether the pre-selected delay time is greater than the upper limit of the delay constraint. If it is, truncate the delay time to the upper limit of the delay constraint and round it down. Otherwise, keep the pre-selected delay time as the delay time. After the screening is completed, reconstruct the one-dimensional micro-fault signal into a multi-dimensional state point sequence. The reconstructed state point sequence is written into the current feature calculation area and stored in association with the average structural element length of the same window, so as to keep the parameters consistent when calculating the subsequent feature offset and discrete rate of change. After forming a multidimensional state point sequence, the state feature analysis module constructs a comparability relationship between the baseline parameter matrix and the current parameter matrix. If only the local dispersion of the current trajectory is compared without considering the offset of the overall distribution center, the sensitivity will be insufficient when the fault characteristics are manifested as an overall trajectory offset. Therefore, in this embodiment, the reference parameter matrix is ​​first decomposed, specifically using the singular value decomposition algorithm, to extract the first few parameters that reflect the main energy distribution characteristics. The set of reference features is composed of 10 maximal singular values, where the parameters are... The method for determining it is as follows: Calculate the sum of all non-zero singular values ​​obtained from matrix decomposition, sort the singular values ​​in descending order and sum them one by one, and select the minimum number of singular values ​​that makes the sum greater than or equal to the product of the sum and a preset energy ratio limit value as the minimum number of singular values. The value, the preset energy ratio limit value, has a constant range. to Between, and calculate this The arithmetic mean of the singular values ​​is used as its reference central quantity; Then, the same singular value decomposition algorithm is used to process the current parameter matrix under the current monitoring window to extract the previous parameters. The current feature set consists of the 100 largest singular values. The values ​​are forced to use the corresponding values ​​determined by the baseline parameter matrix to ensure consistency of the set dimensions, and this is calculated. The arithmetic mean of the singular values ​​is used as its current centrality; The square root of the sum of the squares of the differences between the corresponding features in the current feature set and the benchmark feature set is used to form the feature difference value. Simultaneously, the L2 distance between the current center value and the reference center value is calculated and multiplied by the preset offset weight coefficient to form the center offset value; the feature difference value and the center offset value are normalized and weighted to obtain the feature offset. In this normalized weighted summation process, considering that the feature difference value reflects the change in the shape of the state point group, while the center offset value reflects the change in the overall position, and that the two have different dimensions, the system uses the minimum-maximum linear mapping method to map them to [the appropriate values]. interval; Normalized weighted summation satisfies the following formula: , in, The feature offset is obtained by normalized weighted summation; For characteristic difference values, and These are the predetermined historical minimum and historical range of the characteristic difference values, respectively; This is the center offset value. and These are the predetermined historical minimum and historical range of the center offset value, respectively; and The normalized weight coefficients are preset and satisfy the following conditions: ; The current window contains both changes in the shape of the state distribution and changes in the overall position, both of which are used together for subsequent risk assessment; If the number of valid points in the current window is insufficient during matrix decomposition, the feature offset of the previous valid window will not be directly overwritten. Instead, the window will be marked as a data missing compensation window, and normal calculation will resume when the conditions are met in subsequent windows. When judging solely based on the difference between the current parameter matrix and the baseline parameter matrix, for some early short-pulse anomalies, although the rate of change of the overall parameter matrix is ​​less than the first matrix change threshold, the subsequent evolution between state points will show a divergence trend that precedes the divergence time of the overall parameters. Therefore, it is also necessary to calculate the discrete rate of change. The state feature analysis module filters multiple pairs of state points in the current multidimensional state point sequence whose initial distance is no greater than the preset neighborhood radius; In practice, to avoid the same state point being repeatedly paired with itself or different sampling points having the same value, resulting in an initial distance of zero and causing a calculation anomaly with a zero denominator when calculating the distance ratio, the filtering logic is additionally superimposed with a strict non-zero lower limit constraint. That is, the initial distance of the selected state point pairs must not only be no greater than the preset neighborhood radius, but must also be strictly greater than the preset non-zero lower limit constant, such as the minimum non-zero resolution threshold determined by the least significant bit of the system analog-to-digital converter. For each pair of state points that passes the screening, calculate the ratio of the evolved distance after a preset time step to the initial distance, and perform a logarithmic operation on the ratio to obtain the logarithmic discrete value; The local rate of change of a single state point pair is obtained by dividing the logarithmic discrete value by the product of the time step and the sampling time interval; the local rates of change of all the conditions that meet the requirements in the current window are then summed and divided by the total number of state point pairs to form the discrete rate of change. After the calculation is completed, the discrete rate of change and the aforementioned feature offset are written into the input area of ​​the fault assessment module, and then read by the fault assessment module. If some state point pairs have invalid values ​​in the distance calculation after evolution, only the point pair is removed, without affecting the statistical results of other point pairs; if the number of valid point pairs is lower than the range required for continued statistics, the discrete change rate of the window does not independently trigger new state conclusions, but participates in trend determination together with the window. Under this industrial variable frequency motor, the actual function of the above processing chain is as follows: when the motor load changes, the combined morphological filter first separates the baseline related to the load change from the high-frequency sequence; the phase space reconstruction parameters are not selected independently without the filter scale, but are constrained by the average structural element length. Then, by utilizing the difference between the current parameter matrix and the health reference matrix, as well as the degree of evolutionary separation of neighboring state points, two types of traceable features are formed. In this way, the subsequent modules do not receive a single time-domain waveform, but rather feature offsets and discrete rates of change that have been organized into windows, making it easier to make consistent judgments within the window.

[0021] In a preferred embodiment of the present invention, when the fault assessment module determines the fault risk value based on the feature offset and the discrete change rate, it is specifically used to: read the preset feature offset threshold and discrete change rate threshold from the threshold library corresponding to the equipment model. The fault risk value is obtained by weighting the deviation between the feature offset and the feature offset threshold, and the deviation between the discrete change rate and the discrete change rate threshold. The fault assessment module, when generating fault warning commands or circuit breaker control commands based on changes in fault risk values ​​within the monitoring window, and triggering locking or adjustment operations on at least one of the processing parameters (length, dimension, delay time, and neighborhood radius) of structural elements based on these changes, is specifically used for: The fault risk value corresponding to each monitoring window is calculated using a sliding monitoring window method, and a risk value sequence is constructed; preset normal threshold, early warning threshold and fault deterioration warning threshold are configured, and the normal threshold is less than the early warning threshold, and the early warning threshold is less than the fault deterioration warning threshold. When continuous When the fault risk value within a monitoring window is greater than the early warning threshold but less than or equal to the fault deterioration warning threshold, the electrical equipment is determined to be in an early fault trend, and a fault warning command is generated. It is a natural number greater than 1; the length, dimension, delay time and neighborhood radius of the structure element corresponding to the current monitoring window are locked, and a suspected fault parameter group is formed for archiving; the difference between the feature offset of the current monitoring window and the previous monitoring window is calculated to determine the feature change direction, and the feature offset, discrete change rate, feature change direction and corresponding timestamp of the current monitoring window are written into the pre-established historical record library; When continuous When the fault risk value within a monitoring window is greater than the normal threshold but less than or equal to the early warning threshold, the electrical equipment is determined to be in observation status, and the database status flag is updated. When continuous When the fault risk value within a monitoring window is less than or equal to the normal threshold, the electrical equipment is determined to be in a normal state, and the database status flag is updated. The fault assessment module is also used for: when continuous When the fault risk value in a monitoring window exceeds the fault deterioration warning threshold, the fault is determined to have entered the continuous deterioration stage, and a circuit breaker control command is generated. Extract the most recent normal operation window data and the current abnormal window data before this failure, and write them into a pre-established fault record library in pairs. This allows the fault assessment module to update the feature offset threshold, discrete rate of change threshold, and corresponding feature offset weight and rate of change weight based on the paired data. After the electrical equipment resumes operation, when continuous When the fault risk value within a monitoring window is less than or equal to the normal threshold, the window data under normal conditions is extracted, and the baseline parameter matrix is ​​updated in real time.

[0022] Further explanation: Still targeting the main circuit of an industrial variable frequency motor that frequently starts and stops and is connected to a nonlinear load, the system generates a fault risk value based on the characteristic offset and discrete rate of change. Combined with the threshold configuration of the target equipment model, the current window characteristics and window trends, the system performs status determination and dynamic updates of threshold and benchmark data within the monitoring window. After receiving the feature offset and discrete rate of change from the status feature analysis module, the fault assessment module first reads the feature offset threshold and discrete rate of change threshold corresponding to the device model from the threshold library, and at the same time reads the preset feature offset weight and rate of change weight. If a uniform threshold is used directly without distinguishing the equipment model, the judgment criteria will be inconsistent for motors under different rated operating conditions. Therefore, in this embodiment, both the threshold and the weight are obtained from the records corresponding to the model. The module calculates the ratio of the feature offset to the feature offset threshold, and multiplies this ratio by the feature offset weight to obtain the offset score; then it calculates the ratio of the discrete rate of change to the discrete rate of change threshold, and multiplies this ratio by the rate of change weight to obtain the change score. The offset score and the change score are added together to form the fault risk value of the current monitoring window. After the risk value of the current window is generated, it does not overwrite the original feature record, but is written into the risk value sequence storage area together with the feature offset and discrete change rate of the same window for subsequent window judgment. If the complete configuration of the current device model is not yet available in the threshold library, the current window will only retain the feature data and will not output any new risk conclusions until the threshold configuration is complete and the calculation will resume. When the risk value of a single window fluctuates, issuing an early warning or circuit breaker directly may easily confuse short-term disturbances with persistent faults. Therefore, the fault assessment module calculates the fault risk value corresponding to each monitoring window using a sliding monitoring window method and constructs a risk value sequence; the system pre-configures normal threshold, early warning threshold and fault deterioration warning threshold, and sets the three in order from low to high. For continuous When checking each monitoring window, if the fault risk value is continuously greater than the early warning threshold and less than or equal to the fault deterioration warning threshold, the electrical equipment is determined to be in an early fault trend, and a fault warning instruction is generated. While generating the warning command, the module locks the length of the structural element corresponding to the current monitoring window, sets the dimension, sets the delay time and the neighborhood radius, forms a suspected fault parameter group and archives it. To prevent frequent parameter changes during fault evolution from reducing the comparability of features between windows before and after, locked parameter groups are saved separately and not mixed with general operating parameters for subsequent querying and verification. After identifying an early fault trend, the fault assessment module also calculates the difference in feature offset between the current monitoring window and the previous monitoring window to determine the direction of feature change, and writes the feature offset, discrete rate of change, direction of feature change and corresponding timestamp of the current monitoring window into the historical record database. This record library is used to preserve the state trajectory during the evolution process, facilitating comparison of the fault evolution process after subsequent maintenance; if continuous If the fault risk value within a monitoring window is greater than the normal threshold but less than or equal to the early warning threshold, the device is determined to be in observation mode, and the database status flag is updated. If continuous If the fault risk value within a monitoring window is less than or equal to the normal threshold, the equipment is determined to be in a normal state, and the database status flag is updated. For the observation state, the system only retains continuous tracking and does not perform parameter locking or circuit breaker control. For the normal state, the system continues to evaluate according to subsequent windows. When continuous When the fault risk value within a monitoring window exceeds the fault deterioration warning threshold, the fault assessment module determines that the fault has entered a stage of continuous deterioration and generates a circuit breaker control command. At this time, in addition to sending the circuit breaker control command to the control output module, the most recent normal operation window data and the current abnormal window data before this fault are extracted and written into the fault record database in pairs according to the corresponding relationship. The paired data does not immediately replace the existing threshold, but serves as the data basis for subsequent updates to the feature offset threshold, discrete rate of change threshold, and corresponding feature offset weights and rate of change weights; Paired data is used to make the threshold library correction process based on the corresponding feature differences between normal windows and abnormal windows; The specific execution logic for updating various parameters based on paired data is as follows: The module extracts multiple sets of paired data from the fault record library and calculates the feature offset increment and discrete rate of change increment between the abnormal window and the normal operation window, respectively. The statistical average of the two types of increments is calculated and multiplied by the preset safety margin coefficient to obtain the updated feature offset threshold and discrete rate of change threshold. The specific threshold update algebraic mapping relationship satisfies the formula: , in, For the updated feature offset threshold or discrete rate of change threshold, This corresponds to the original threshold before the update. For safety margin coefficient, The number of paired data sets extracted. For the first The component consists of the feature offset increment or discrete rate of change increment calculated from the data. The calibration basis for this safety margin coefficient is derived from the rated insulation withstand limit of the corresponding motor model and the time lead required for historical samples. This is typically... The threshold is selected from among several values ​​to ensure that the new threshold also takes into account the margin for disturbance rejection; At the same time, the relative magnitude proportions of the two types of increments are compared, and the corresponding feature offset weights and rate of change weights are dynamically adjusted according to the inverse proportional rule. Since the feature offset increment belonging to the normalized scale and the discrete rate of change increment belonging to the reciprocal time scale have completely different physical dimensions, directly adding the two values ​​will cause serious dimensional conflicts and proportional distortion. Therefore, the specific quantization execution logic of this inverse proportional rule is optimized as follows: divide the calculated feature offset increment by the feature offset threshold before the update, and convert it into a dimensionless offset relative increment rate. At the same time, the discrete rate of change increment is divided by the discrete rate of change threshold before the update to convert it into a dimensionless relative rate of change. Based on the unification of dimensions, the first proportion of the relative increment rate of the offset in the sum of the two dimensionless increment rates and the second proportion of the relative increment rate of change in the sum of the two dimensionless increment rates are calculated respectively. The new feature offset weight is assigned to the second proportion, and the new rate of change weight is assigned to the first proportion. Thus, through cross-mapping under a unified dimension, feature items with relative increments lower than the mean and trigger probabilities lower than the preset probability are given a judgment weight that increases their value. If the interruption device has not completed the disconnection feedback during the continuous deterioration phase, the system will retain the current abnormal window and the risk sequence of subsequent windows and will not discard the fault records that have been formed. After the equipment resumes operation, the system does not directly incorporate the first low-risk window after recovery into the baseline parameter matrix; it only incorporates it when consecutive low-risk windows are present. Only when the fault risk value within a monitoring window is less than or equal to the normal threshold again will the window data under normal conditions be extracted and the benchmark parameter matrix be updated in real time. This avoids including unstable transition window data in the baseline reference; when updating the baseline parameter matrix, only data that has recovered to the normal state and meets the window conditions are used, and the original fault window data is retained in the fault record library and does not participate in the health baseline update; Therefore, under this industrial variable frequency motor, the fault assessment module not only completes the calculation of the current window risk value, but also completes window status identification, early warning and circuit breaker output, parameter group archiving, historical record writing, fault sample accumulation, and baseline update after recovery.

[0023] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A pre-diagnosis-based equipment fault prediction system, characterized in that, include: The data acquisition module is used to acquire the high-frequency current or voltage time series of the main circuit of the electrical equipment, and simultaneously calculate the real-time variable load current gradient sequence of the main circuit. The morphological filtering module is used to determine the length of the structural element based on the real-time variable load current gradient sequence, filter the high-frequency current or voltage time sequence using a morphological combined filter with the length of the structural element to obtain the load baseline, and perform a differential operation between the load baseline and the high-frequency current or voltage time sequence to obtain a minor fault signal. The state characteristic analysis module is used to establish a reference parameter matrix based on the high-frequency current or voltage time series during normal operation; Determine the dimension and delay time of the phase space reconstruction, and reconstruct the phase space of the minute fault signal according to the dimension and the delay time, mapping it into a multi-dimensional state point sequence; The difference between the current parameter matrix of the sequence and the reference parameter matrix under the current monitoring window is calculated to obtain the feature offset; And track the distance changes of multiple state point pairs in the sequence whose initial distance is not greater than the preset neighborhood radius after a preset time step to obtain the discrete rate of change; The fault assessment module is used to determine the fault risk value based on the feature offset and the discrete change rate, and generate a fault warning command or circuit breaker control command according to its change in the monitoring window, and trigger a locking or adjustment operation on at least one of the processing parameters, namely the length of the structural element, the dimension, the delay time and the preset neighborhood radius. The control output module is used to send the fault warning command and the circuit breaker control command to the alarm terminal and the circuit breaker execution device, respectively.

2. The equipment fault prediction system based on pre-diagnosis according to claim 1, characterized in that, The data acquisition module is specifically used for: Collect the power frequency main circuit current sampling value of the main circuit; Extract the variation characteristics of the main circuit current sample values ​​at adjacent sampling times; based on the correlation between the variation characteristics and the sampling time interval, determine the real-time variable load current gradient sequence at the corresponding sampling time.

3. The equipment fault prediction system based on pre-diagnosis according to claim 1, characterized in that, When determining the length of the structural element based on the real-time variable load current gradient sequence, the morphological filtering module is specifically used for: Obtain the basic structural element length, rated maximum current gradient threshold, filter attenuation coefficient, and minimum structural element length lower limit of the electrical equipment; The attenuation factor is determined based on the mapping relationship between the real-time variable load current gradient sequence, the rated maximum current gradient threshold, and the filter attenuation coefficient. The length of the candidate structural element is determined based on the length of the basic structural element and the attenuation factor. The maximum value between the candidate structuring element length and the minimum structuring element length lower limit is selected as the structuring element length.

4. The equipment fault prediction system based on pre-diagnosis according to claim 1, characterized in that, When the morphological filtering module uses a morphological combination filter with the length of the structuring element to filter the high-frequency current or voltage time series to obtain the load baseline, it is specifically used for: A flat structural element is constructed based on the length of the structural element; the flat structural element is used to sequentially perform morphological opening and morphological closing operations on the high-frequency current or voltage time series to obtain an opening-closing filter sequence; Using the flat structural element constructed based on the length of the structural element, morphological closing and morphological opening operations are sequentially performed on the high-frequency current or voltage time series to obtain a closed-open filter sequence; The load baseline is obtained by adding the open-closed filter sequence to the closed-open filter sequence and dividing by two.

5. The equipment fault prediction system based on pre-diagnosis according to claim 1, characterized in that, When the state feature analysis module reconstructs the phase space of the minute fault signal and maps it into a multi-dimensional state point sequence, it is specifically used for: The pre-selected dimension is determined by the nearest neighbor method, and the pre-selection delay time is determined by correlation analysis. Obtain the length of the structuring element output by the morphological filtering module under the current monitoring window, and calculate the average structuring element length; Based on the average structural element length, the pre-selected dimension and the pre-selected delay time are constrained and filtered according to a preset proportional relationship to determine the dimension and the delay time; According to the dimension and the delay time, the one-dimensional minute fault signal is reconstructed into the multi-dimensional state point sequence.

6. The equipment fault prediction system based on pre-diagnosis according to claim 1, characterized in that, When the state feature analysis module calculates the difference between the current parameter matrix and the reference parameter matrix of the multidimensional state point sequence under the current monitoring window to obtain the feature offset, it is specifically used for: Feature extraction is performed on the baseline parameter matrix and the current parameter matrix under the current monitoring window to obtain the baseline feature set and the current feature set and their corresponding center quantities; Based on the degree of numerical deviation between the current feature set and the reference feature set, and the spatial offset relationship between the corresponding center quantities, the feature offset is calculated by feature fusion.

7. The equipment fault prediction system based on pre-diagnosis according to claim 1, characterized in that, When the state feature analysis module tracks the distance change between two state points in the multidimensional state point sequence whose initial distance is no greater than a preset neighborhood radius after a preset time step and obtains the discrete rate of change, it is specifically used for: In the multidimensional state point sequence, multiple pairs of state points with an initial distance not greater than the neighborhood radius are selected; Local change features are extracted based on the ratio of the distance between each state point pair after the time step to the initial distance; the discrete change rate is obtained by combining the local change features of multiple state point pairs.

8. The equipment fault prediction system based on pre-diagnosis according to claim 1, characterized in that, When determining the fault risk value based on the feature offset and the discrete rate of change, the fault assessment module is specifically used for: Read the preset feature offset threshold and discrete change rate threshold from the threshold library corresponding to the device model; The fault risk value is obtained by weighting the deviations of the feature offset from the feature offset threshold and the deviations of the discrete change rate from the discrete change rate threshold.

9. The equipment fault prediction system based on pre-diagnosis according to claim 1, characterized in that, When the fault assessment module generates a fault warning command or a circuit breaker control command based on the changes in the fault risk value within the monitoring window, and triggers a locking or adjustment operation on at least one processing parameter among the structural element length, the dimension, the delay time, and the neighborhood radius based on the changes, it is specifically used for: The fault risk value corresponding to each monitoring window is calculated using a sliding monitoring window method, and a risk value sequence is constructed. Configure a preset normal threshold, an early warning threshold, and a fault deterioration warning threshold, wherein the normal threshold is less than the early warning threshold, and the early warning threshold is less than the fault deterioration warning threshold; When continuous When the fault risk value within a monitoring window is greater than the early warning threshold and less than or equal to the fault deterioration warning threshold, the electrical equipment is determined to be in an early fault trend, and the fault warning instruction is generated. The value is a natural number greater than 1; the length of the structural element, the dimension, the delay time, and the neighborhood radius corresponding to the current monitoring window are locked to form a suspected fault parameter group for archiving; the difference between the feature offset of the current monitoring window and the previous monitoring window is calculated to determine the feature change direction, and the feature offset, the discrete change rate, the feature change direction, and the corresponding timestamp of the current monitoring window are written into a pre-established historical record library; When continuous When the fault risk value in a monitoring window is greater than the normal threshold and less than or equal to the early warning threshold, the electrical equipment is determined to be in observation status, and the database status flag is updated. When continuous When the fault risk value within a monitoring window is less than or equal to the normal threshold, the electrical equipment is determined to be in a normal state, and the database status flag is updated.

10. The equipment fault prediction system based on pre-diagnosis according to claim 9, characterized in that, The fault assessment module is also used for: When continuous When the fault risk value in a monitoring window is greater than the fault deterioration warning threshold, the fault is determined to have entered a continuous deterioration stage, and the circuit breaker control command is generated. Extract the most recent normal operating window data and the current abnormal window data before this failure, and write them into a pre-established fault record library in pairs. This allows the fault assessment module to update the feature offset threshold, discrete rate of change threshold, and corresponding feature offset weight and rate of change weight based on the paired data. After the electrical equipment resumes operation, when continuously When the fault risk value within a monitoring window is less than or equal to the normal threshold, the window data under normal conditions is extracted, and the benchmark parameter matrix is ​​updated in real time.

Images