A method and system for monitoring the status of relays in a metro train

Abstract

The present application relates to the technical field of electrical fault location, in particular to a kind of urban rail train relay state online monitoring method and system.The specific implementation process includes: using electrical sensor group to collect coil current, contact voltage, load current and environmental vibration signal;When coil current occurs step change, intercepts electromechanical characteristic data set;Calculate arc energy and contact resistance fluctuation characteristics, correlation analysis is carried out in combination with environmental vibration signal, distinguish passive fluctuation and active fluctuation;Establish multi-factor health degree evaluation model, according to the size of load current and vibration intensity dynamically adjusts the weight coefficient of arc energy and contact resistance, real-time monitoring relay health state and predicting remaining life.The present application is through multi-dimensional data synchronous acquisition and vibration decoupling, effectively eliminates the interference of mechanical vibration to electrical performance monitoring, improves the problem that offline test cannot reproduce dynamic working condition, significantly improves the accuracy of relay state evaluation and life prediction.

Description

Technical Field

[0001] This invention relates to the field of electrical fault location technology, specifically to a method and system for online monitoring of the status of relays in urban rail trains. Background Technology

[0002] In the field of electrical fault location technology, relays are commonly used in urban rail transit vehicles as logic control and signal isolation components. Their electrical performance directly affects the reliable operation of train traction, braking, door control, and safety circuits. Existing technologies mainly rely on fixed-cycle planned maintenance or offline electrical performance testing. This involves manually measuring static electrical parameters such as coil impedance, contact resistance, insulation resistance, and pull-in / release voltage of the relays one by one using general measuring instruments such as multimeters after the train returns to the depot, or by removing the relays from the car body cabinet and placing them on a dedicated offline test bench.

[0003] However, existing technologies have inherent limitations in practical applications. Because offline testing environments are typically static and idealized, they cannot realistically reproduce the dynamic conditions experienced by trains during high-speed operation, such as severe mechanical vibrations, wide-range temperature variations, strong electromagnetic interference, and complex load fluctuations. This can lead to relays that pass static electrical parameter testing intermittently failing during actual energized operation due to unstable dynamic contact resistance or momentary arcing, resulting in a significant disconnect between test results and actual operating conditions. In summary, existing technologies have long-term blind spots in condition monitoring between maintenance intervals, failing to capture in real-time degradation characteristics of relay electrical performance (such as pull-in time drift and accelerated contact oxidation) over time. Consequently, they lack the ability to detect sudden electrical faults in real time and predict trends based on actual health conditions.

[0004] To address this, a method and system for online monitoring of the status of relays in urban rail transit trains are proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for online monitoring of the status of relays in urban rail trains, so as to monitor the status of relays in urban rail trains online.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for online monitoring of the status of relays in urban rail transit trains includes:

[0008] Using an electrical sensor array installed on an urban rail train, the coil current, contact voltage, load current, and environmental vibration signals of the relay are collected; when the coil current undergoes a step change, the electromechanical characteristic data set within the time window before and after the change is automatically locked and synchronously captured.

[0009] During contact operation, the arcing phenomenon of contact voltage and load current is monitored, and the product of voltage and current during the arcing period is integrated over time to calculate the arc energy; under the stable engagement state, the contact resistance and fluctuation characteristics of the contact are calculated, and the correlation analysis of the electromechanical characteristic dataset is performed in conjunction with the environmental vibration signal to distinguish between passive fluctuations caused by vibration and active fluctuations caused by contact deterioration.

[0010] A multi-factor health assessment model is established based on the arc energy, active fluctuations, and passive fluctuations. The weighting coefficients of arc energy and contact resistance in the health assessment are dynamically adjusted according to the load current and vibration signal intensity at the time of contact action. The assessed single damage value is accumulated into the historical damage value to monitor the health status of the relay in real time and predict the remaining life.

[0011] Preferably, the specific implementation process of using an electrical sensor array installed on an urban rail train to collect relay coil current, contact voltage, load current, and environmental vibration signals includes:

[0012] An electrical sensor group is configured on the relay control circuit to continuously sense the magnetic field changes generated by the excitation of the coil and generate a load current signal; the load current signal reflecting the state of the controlled equipment is captured in the execution circuit to generate a digital sequence vector of coil current, contact voltage, load current and environmental vibration with timestamp.

[0013] Preferably, the specific implementation process of automatically locking and synchronously capturing the electromechanical feature data set within the time window before and after the change when the coil current undergoes a step change includes:

[0014] A sliding difference operation is performed on the digital sequence vector to calculate the amplitude change rate between adjacent sampling points, and the amplitude change rate is logically compared with the action trigger threshold. When the amplitude change rate exceeds the action trigger threshold, it is determined that a step change has occurred and a time anchor signal is generated. The time anchor signal is time-aligned and the transmission delay difference is eliminated. The aligned multidimensional time series is encapsulated into an electromechanical feature dataset containing transient and steady-state processes.

[0015] Preferably, during contact operation, the arcing phenomenon of contact voltage and load current is monitored, and the product of voltage and current during arcing is integrated over time to calculate the arc energy. The specific implementation process includes:

[0016] The contact voltage sequence and load current sequence are extracted from the electromechanical feature dataset. The first-order differential operation is performed on the contact voltage sequence to identify the transient interval where the voltage amplitude changes drastically and to locate the time range of contact action. The numerical integration algorithm is applied to accumulate and sum the voltage and current of the arc phenomenon within the time range of contact action to calculate the arc energy characterizing the degree of contact ablation.

[0017] Preferably, under the stable engagement state, the specific implementation process for calculating the contact resistance and fluctuation characteristics of the contacts includes:

[0018] The starting point of the steady-state engagement is determined by the amplitude convergence of the contact voltage waveform. The contact voltage sequence and load current sequence after the starting point are extracted from the electromechanical feature dataset, and point-by-point synchronous division is performed to generate an instantaneous contact resistance data stream containing time-varying characteristics. A statistical smoothing filtering algorithm is used to remove outlier noise points in the instantaneous contact resistance data stream to obtain the contact resistance. At the same time, the discreteness analysis of the instantaneous contact resistance data stream is performed to calculate the variance and standard deviation statistics and extract the fluctuation characteristics that characterize the stability of the contact resistance.

[0019] Preferably, the specific implementation process of performing correlation analysis on the electromechanical feature dataset in conjunction with the environmental vibration signal to distinguish between passive fluctuations caused by vibration and active fluctuations caused by contact deterioration includes:

[0020] The environmental vibration time-domain waveform and instantaneous fluctuation sequence of contact resistance are retrieved from the electromechanical feature dataset. Frequency normalization is performed using a polynomial interpolation algorithm to establish a time-axis aligned synchronous analysis vector. The correlation coefficient flow in the time domain is calculated using a sliding window cross-correlation function to identify and lock the time segments exhibiting correlation. The transfer function is established by fitting using the least squares method to estimate the resistance change component directly excited by mechanical vibration and label it as passive fluctuation. The passive fluctuation is subtracted from the instantaneous fluctuation of contact resistance to obtain the residual sequence after removing external mechanical interference. The stationarity test and trend analysis are performed on the residual sequence to extract the intrinsic change characteristics reflecting the physical wear, electrochemical corrosion, and contact surface ablation of the contact material and label them as active fluctuation.

[0021] Preferably, the specific implementation process of establishing a multi-factor health assessment model based on the arc energy, active fluctuations, and passive fluctuations includes:

[0022] The system receives the arc energy, active fluctuations, and passive fluctuations. It then uses the relay's rated parameter set to normalize the arc energy and active fluctuations, generating electrical damage factor data and mechanical damage factor data. For the passive fluctuations, it applies nonlinear mapping logic to generate operating condition correction factor data reflecting the degree of external environmental interference. A multi-dimensional state feature space is constructed, mapping the electrical damage factor data, mechanical damage factor data, and operating condition correction factor data into state vectors within the feature space. These state vectors are then set as input variables to establish a multi-factor health assessment model.

[0023] Preferably, the specific implementation process of dynamically adjusting the weighting coefficients of arc energy and contact resistance in the health assessment based on the load current magnitude and vibration signal intensity during the current contact operation includes:

[0024] The relay load characteristic curve and vibration grading threshold table are called, and the load current magnitude is converted into a load stress index and the vibration signal intensity is converted into a mechanical interference index using a nonlinear mapping function. A two-dimensional weighted decision matrix based on fuzzy logic is constructed. The load stress index and mechanical interference index are used as index variables to query and calculate the proportion of damage contribution of arc energy and contact resistance under the current working condition, and load it into the parameter configuration port of the multi-factor health assessment model.

[0025] Preferably, the specific implementation process of accumulating the assessed single damage value into the historical damage level to monitor the health status of the relay in real time and predict its remaining life includes:

[0026] The historical damage data from the previous monitoring period is read and superimposed with the single damage value output by the multi-factor health assessment model to generate updated current cumulative damage data. The current cumulative damage data is compared with the relay failure threshold curve to calculate the proportion of the current damage level in the total lifespan capacity. Recent historical data segments are retrieved, and regression analysis algorithms are used to calculate the slope parameter of the damage growth trend. The current cumulative damage data is extrapolated to the future trend to calculate the estimated number of actions required to reach the failure threshold.

[0027] An online monitoring system for the status of relays in urban rail transit includes:

[0028] The multi-dimensional data acquisition module uses an electrical sensor array installed on the urban rail train to collect relay coil current, contact voltage, load current, and environmental vibration signals; when the coil current undergoes a step change, it automatically locks and synchronously captures the electromechanical feature data set within the time window before and after the change.

[0029] The characteristic fluctuation differentiation module monitors the arcing phenomenon of contact voltage and load current during contact operation, performs time integration on the product of voltage and current during arcing, and calculates the arc energy; in the stable state of engagement, it calculates the contact resistance and fluctuation characteristics of the contact, and performs correlation analysis on the electromechanical characteristic dataset in conjunction with the environmental vibration signal to distinguish between passive fluctuations caused by vibration and active fluctuations caused by contact deterioration.

[0030] The health monitoring module establishes a multi-factor health assessment model based on the arc energy, active fluctuations, and passive fluctuations. It dynamically adjusts the weighting coefficients of arc energy and contact resistance in the health assessment according to the load current magnitude and vibration signal intensity at the time of contact action. It accumulates the assessed single damage value into the historical damage value, monitors the health status of the relay in real time, and predicts the remaining life.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] 1. This invention utilizes an electrical sensor array to collect data during high-speed train operation. This enables the system to capture intermittent faults (such as unstable dynamic contact resistance or instantaneous arcing) that are only exposed under real dynamic operating conditions such as severe mechanical vibration, wide-range temperature changes, and complex load fluctuations. This solves the problem of the disconnect between offline test results and actual operating conditions.

[0033] 2. This invention effectively distinguishes between "passive fluctuations" caused by mechanical vibration and "active fluctuations" caused by physical wear / corrosion of contact materials by collecting environmental vibration signals and performing correlation analysis with electromechanical characteristic data. This method uses the least squares method to fit the transfer function to filter out external mechanical interference, ensuring the accuracy of judging the true degree of degradation (contact resistance stability) of relay contacts.

[0034] 3. This invention establishes a multi-factor health assessment model and dynamically adjusts the weights based on the load current magnitude and vibration intensity. By accumulating single damage values ​​to historical damage values ​​and combining them with regression analysis algorithms, the system can not only monitor the current health status in real time, but also extrapolate future degradation trends and predict remaining lifespan, thus realizing online monitoring of the relay status of urban rail trains. Attached Figure Description

[0035] Figure 1 This is a flowchart of an online monitoring method for the status of relays in urban rail trains proposed in this invention;

[0036] Figure 2 This is a structural diagram of an online monitoring system for the status of relays in urban rail trains proposed in this invention;

[0037] Figure 3 This is a schematic diagram of the relay status monitoring process proposed in this invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It must be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to constitute any limitation on the scope of protection of this invention. Therefore, all equivalent changes or modifications conceived by those skilled in the art based on the content disclosed in this invention without inventive effort should fall within the scope of protection claimed by this invention.

[0039] Reference Figures 1 to 3 This invention proposes an online monitoring method and system for the status of relays in urban rail trains, the technical solution of which is as follows:

[0040] Example 1:

[0041] Reference Figure 1 This embodiment proposes an online monitoring and inspection method for the status of relays in urban rail transit trains, including:

[0042] Using an electrical sensor array installed on an urban rail train, the coil current, contact voltage, load current, and environmental vibration signals of the relay are collected; when the coil current undergoes a step change, the electromechanical characteristic data set within the time window before and after the change is automatically locked and synchronously captured.

[0043] During contact operation, the arcing phenomenon of contact voltage and load current is monitored, and the product of voltage and current during the arcing period is integrated over time to calculate the arc energy; under the stable engagement state, the contact resistance and fluctuation characteristics of the contact are calculated, and the correlation analysis of the electromechanical characteristic dataset is performed in conjunction with the environmental vibration signal to distinguish between passive fluctuations caused by vibration and active fluctuations caused by contact deterioration.

[0044] A multi-factor health assessment model is established based on the arc energy, active fluctuations, and passive fluctuations. The weighting coefficients of arc energy and contact resistance in the health assessment are dynamically adjusted according to the load current and vibration signal intensity at the time of contact action. The assessed single damage value is accumulated into the historical damage value to monitor the health status of the relay in real time and predict the remaining life.

[0045] Furthermore, the specific implementation process of using electrical sensor arrays installed on urban rail trains to collect relay coil current, contact voltage, load current, and environmental vibration signals includes:

[0046] An electrical sensor group is configured on the relay control circuit to continuously sense the magnetic field changes generated by the excitation of the coil and generate a load current signal; the load current signal reflecting the state of the controlled equipment is captured in the execution circuit to generate a digital sequence vector of coil current, contact voltage, load current and environmental vibration with timestamp.

[0047] Specifically, an electrical sensor group is configured on the relay control circuit, comprising a Hall current sensor, a non-contact voltage probe, and a MEMS accelerometer. A Hall current sensor with a range of 0-200mA is snap-fitted to the coil input terminal of the relay to continuously sense changes in the magnetic field generated by the coil excitation and generate a coil current signal reflecting the control side status. A Hall current sensor with a range of 0-20A is connected in series in the relay's execution circuit (i.e., the contact load circuit), and a high-impedance voltage acquisition probe is connected in parallel across the contacts to capture the load current signal and contact voltage drop signal reflecting the controlled equipment status, respectively. A MEMS accelerometer with a sampling frequency set to 2000Hz is bonded to a mounting backplate near the relay base to collect environmental vibration signals.

[0048] The sampling clock frequency is set to 100kHz, and the acquired signals are digitized using a multi-channel synchronous data acquisition card. When the coil control circuit is energized, the Hall sensor senses the change in magnetic flux and outputs the corresponding analog voltage, which is then converted into a digitized coil current data stream after A / D conversion. The sensors in the execution circuit record the waveform changes of the load current and contact voltage in real time. An FPGA hardware clock is used to stamp the data of all channels with a unified timestamp, eliminating the slight delay differences caused by the transmission lines of each sensor. The coil current value (e.g., 45mA in steady state), contact voltage value (e.g., 15mV after pull-in), load current value (e.g., 8.5A), and environmental vibration acceleration value (e.g., 0.2g) at the same moment are combined to generate a digital sequence vector of coil current, contact voltage, load current, and environmental vibration with timestamps.

[0049] This embodiment, through multi-dimensional data synchronous acquisition and encapsulation, places mechanical vibration signals and electrical performance signals under the same time reference. This not only improves the problem that offline testing cannot reproduce the dynamic operating conditions of trains, but also effectively establishes the time-domain correspondence between mechanical disturbances and electrical parameter fluctuations, thereby enhancing the anti-interference capability of relay status monitoring in complex electromagnetic and mechanical environments.

[0050] Furthermore, the specific implementation process of automatically locking and synchronously capturing the electromechanical characteristic data set within the time window before and after the change when the coil current undergoes a step change includes:

[0051] A sliding difference operation is performed on the digital sequence vector to calculate the amplitude change rate between adjacent sampling points, and the amplitude change rate is logically compared with the action trigger threshold. When the amplitude change rate exceeds the action trigger threshold, it is determined that a step change has occurred and a time anchor signal is generated. The time anchor signal is time-aligned and the transmission delay difference is eliminated. The aligned multidimensional time series is encapsulated into an electromechanical feature dataset containing transient and steady-state processes.

[0052] Specifically, a sliding differential operation is performed on the real-time input digital sequence vector of the coil current. By calculating the difference between the current sampling point value and the previous sampling point value and dividing it by the sampling time interval, the amplitude change rate, which characterizes the transient change rate of the current signal, is obtained in real time. Taking a certain model of DC110V train relay as an example, considering that the current rise slope is extremely high at the moment the coil circuit is closed, the action trigger threshold is set to 50mA / ms. The real-time calculated amplitude change rate is logically compared with the action trigger threshold. At the moment the control power is turned on in the relay coil circuit, the coil current will rise rapidly from 0mA to the rated operating current (e.g., 45mA). Once the amplitude change rate is detected to exceed the set threshold of 50mA / ms, it is determined that a step change has occurred on the relay control side, and a time anchor signal with a time stamp is immediately generated, marking this moment as the zero moment (T0) of the event.

[0053] Due to the potential microsecond-level transmission delay differences caused by different types of sensors (such as Hall current sensors and MEMS vibration sensors) and varying signal transmission cable lengths, it is necessary to perform time-series alignment processing on the time anchor signal. The specific processing procedure is as follows: Based on the hardware delay parameters of each channel (e.g., 5μs delay for the current channel and 12μs delay for the vibration channel), and using the time anchor signal as a reference, corresponding time axis shift compensation is performed on the data of each channel to eliminate transmission delay differences and ensure that all physical quantities are strictly aligned on the time axis. On this basis, taking time T0 as the center, 50ms of historical data is extracted to the left as a steady-state background reference (including background noise and initial state), and 450ms of real-time data is extracted to the right as a record of the action process (including contact closure, arc oscillation, and pull-in stabilization process). This total of 500ms of time-aligned coil current, contact voltage, load current, and environmental vibration data segments are uniformly encapsulated into an electromechanical feature dataset containing both transient and steady-state processes.

[0054] This embodiment, through step detection based on amplitude change rate and sliding window extraction, can extract only effective data segments containing the instantaneous relay action and its preceding and following correlations from the large amount of redundant data generated by train operation, greatly reducing the load on data storage and processing. By using time anchor point generation based on electrical signal step and a multi-channel timing alignment mechanism, the timing error caused by sensor hardware characteristics is effectively eliminated, ensuring the synchronization of electrical characteristics (such as arcing) and mechanical characteristics (such as vibration).

[0055] Furthermore, during contact operation, the arcing phenomenon of contact voltage and load current is monitored, and the product of voltage and current during arcing is integrated over time to calculate the specific realization process of arc energy, including:

[0056] The contact voltage sequence and load current sequence are extracted from the electromechanical feature dataset. The first-order differential operation is performed on the contact voltage sequence to identify the transient interval where the voltage amplitude changes drastically and to locate the time range of contact action. The numerical integration algorithm is applied to accumulate and sum the voltage and current of the arc phenomenon within the time range of contact action to calculate the arc energy characterizing the degree of contact ablation.

[0057] Specifically, time-axis aligned contact voltage and load current sequences are extracted from the electromechanical feature dataset. Since arcing is typically accompanied by severe oscillations and abrupt changes in voltage waveforms, a first-order differential operation needs to be performed on the contact voltage sequence. The specific process of this first-order differential operation is as follows: the ratio of the voltage difference between adjacent sampling points to the sampling time interval is calculated sequentially to obtain a slope sequence reflecting the rate of voltage change. At the instant the relay contacts separate, the voltage between the contacts is no longer a near-zero conduction voltage drop, but rather rises rapidly and may generate an arc voltage capable of breaking down the air. When the voltage slope sequence is detected to exceed the differential threshold for the first time... The moment when the voltage slope returns to zero and the voltage amplitude stabilizes at the power supply voltage or falls back to zero (successful engagement) is marked as the arc initiation point; the moment when the voltage slope returns to zero and the voltage amplitude stabilizes at the power supply voltage or falls back to zero is marked as the arc termination point. The differential threshold... The calculation formula is as follows:

[0058] ;

[0059] in This represents the mean of the background noise samples; The standard deviation of the background noise sample is used as the background noise sample in this embodiment, which is the differential sequence of the contact voltage 2ms before the contact action. For safety reasons, in this embodiment, we take... By identifying transient intervals where voltage amplitude changes drastically, the timing of contact action and arcing duration can be precisely determined. For example, the arcing interval of this action can be locked as 10.5 milliseconds to 12.2 milliseconds, with a duration of 1.7 milliseconds.

[0060] After determining the arcing time range, a numerical integration algorithm is applied to calculate the energy within that range. Using the trapezoidal integration rule or the rectangular accumulation method, the contact voltage sample value (e.g., arcing voltage 25V) and load current sample value (e.g., load current 8.5A) are read point by point within the locked time range at the same instant. These two values ​​are multiplied to obtain the instantaneous arc power, which is then multiplied by the sampling time interval (e.g., 10μs). The instantaneous energy increments at all sampling points within this time range are accumulated and summed to finally calculate the arc energy characterizing the degree of contact ablation. For example, after accumulating and calculating thousands of data points within the aforementioned 1.7 milliseconds, the arc energy generated by this action is found to be 15.3 mJ.

[0061] This embodiment, by performing first-order differential processing on the voltage waveform, can capture the start and end times of arcing from a complex electromagnetic interference background, improving the time positioning drift problem caused by relying solely on current thresholds to determine arcing. The numerical integration algorithm using the product of voltage and current is used to calculate the arc energy, which, compared to simply recording the number of actions or arcing duration, more realistically reflects the degree of thermal shock and physical ablation experienced by the contact material during each action.

[0062] Furthermore, under the stable engagement state, the specific implementation process for calculating the contact resistance and fluctuation characteristics of the contacts includes:

[0063] The starting point of the steady-state engagement is determined by the amplitude convergence of the contact voltage waveform. The contact voltage sequence and load current sequence after the starting point are extracted from the electromechanical feature dataset, and point-by-point synchronous division is performed to generate an instantaneous contact resistance data stream containing time-varying characteristics. A statistical smoothing filtering algorithm is used to remove outlier noise points in the instantaneous contact resistance data stream to obtain the contact resistance. At the same time, the discreteness analysis of the instantaneous contact resistance data stream is performed to calculate the variance and standard deviation statistics and extract the fluctuation characteristics that characterize the stability of the contact resistance.

[0064] Specifically, time-domain analysis is performed on the contact voltage waveforms in the electromechanical feature dataset. A sliding window algorithm is used to detect the convergence of the voltage amplitude to determine the starting point of the steady-state pull-in. In the initial stage of contact closure, a mechanical bounce typically lasts for about 1-5 milliseconds, during which the voltage oscillates between zero volts and the power supply voltage. When the contact voltage amplitude remains below the convergence threshold (e.g., 50mV) for a continuous 20-millisecond time window, the mechanical bounce is considered to have ended, and the start of this time window is marked as the starting point of the steady-state pull-in. The contact voltage sequence and load current sequence for the interval from this starting point until the disconnect command is issued are extracted from the electromechanical feature dataset. A point-by-point synchronous division operation is performed on these two time-aligned sequences, i.e., the contact voltage drop value at each moment is divided by the load current value, thereby generating the instantaneous contact resistance data stream containing time-varying characteristics.

[0065] Because the electromagnetic environment at the site may introduce high-frequency noise into the measurement circuit, causing unrealistic spikes in the original data stream, a statistical smoothing filtering algorithm is used to remove outlier noise points from the instantaneous contact resistance data stream. Specifically, a moving median filter with a window width of 50 sampling points is applied to smooth the data stream and filter out abrupt noise caused by electromagnetic interference. The arithmetic mean of the filtered data stream is determined as the contact resistance value for that action (e.g., the final calculated static contact resistance is 12.5 milliohms). To capture the dynamic stability of the contact points under train vibration, a dispersion analysis is performed on the unsmoothed or only extreme noise-removed instantaneous contact resistance data stream. By calculating the variance and standard deviation statistics of the data stream, the degree of resistance value fluctuation is quantified. For example, if the calculated standard deviation of the contact resistance is only 0.2 milliohms, it indicates a tight and stable contact; if the standard deviation is as high as 5 milliohms, even if the average resistance is acceptable, it indicates looseness or microscopic contact defects between the contacts. These statistics are extracted as fluctuation characteristics characterizing the stability of the contact resistance.

[0066] This embodiment achieves the capture of dynamic contact resistance by establishing a steady-state determination mechanism based on voltage amplitude convergence and a dynamic resistance current calculation method. By introducing statistical indicators such as variance and standard deviation to quantify resistance fluctuations, it is possible to identify early fault symptoms where the contact pressure is unstable at the microscopic level due to vibration or oxide film, even if the contact is not completely disconnected, thus significantly improving the reliability of relay monitoring.

[0067] Furthermore, the specific implementation process of performing correlation analysis on the electromechanical feature dataset based on the environmental vibration signals to distinguish between passive fluctuations caused by vibration and active fluctuations caused by contact deterioration includes:

[0068] The environmental vibration time-domain waveform and instantaneous fluctuation sequence of contact resistance are retrieved from the electromechanical feature dataset. Frequency normalization is performed using a polynomial interpolation algorithm to establish a time-axis aligned synchronous analysis vector. The correlation coefficient flow in the time domain is calculated using a sliding window cross-correlation function to identify and lock the time segments exhibiting correlation. The transfer function is established by fitting using the least squares method to estimate the resistance change component directly excited by mechanical vibration and label it as passive fluctuation. The passive fluctuation is subtracted from the instantaneous fluctuation of contact resistance to obtain the residual sequence after removing external mechanical interference. The stationarity test and trend analysis are performed on the residual sequence to extract the intrinsic change characteristics reflecting the physical wear, electrochemical corrosion, and contact surface ablation of the contact material and label them as active fluctuation.

[0069] Specifically, the environmental vibration time-domain waveform and instantaneous fluctuation sequence of contact resistance are retrieved from the electromechanical feature dataset. The environmental vibration data with a low sampling rate is upsampled and frequency normalized using a polynomial interpolation algorithm (e.g., cubic spline interpolation) to ensure its time resolution matches that of the contact resistance data, thus establishing a synchronous analysis vector with a strictly aligned time axis. Based on this, the correlation coefficient flow in the time domain is calculated using a sliding window cross-correlation function. A time window length (e.g., 50 milliseconds) is set, and this window slides along the synchronous analysis vector at a set step size, calculating the cross-correlation coefficient between vibration amplitude changes and resistance value changes within the window in real time. Taking a scenario where a train experiences a severe impact in a switch area as an example, it is detected that the cross-correlation coefficient remains consistently above the strong correlation threshold (e.g., 0.75) within a 200-millisecond time period. This indicates a high degree of synchronicity between resistance fluctuations and mechanical vibration during this period, thus identifying this highly correlated time segment as the key analysis interval. Within the locked time segment, the least squares method is used to perform regression fitting on the vibration input and resistance output, establishing a transfer function reflecting the conversion relationship from mechanical force to contact resistance. This transfer function quantitatively describes the theoretical resistance drift caused by changes in contact pressure at the contacts under vibration excitation of a specific frequency and amplitude in the current relay structure. Specifically, the transfer function employs a discrete-time difference equation incorporating time-delay characteristics, as shown in the following formula:

[0070] ;

[0071] in These are the weighting coefficients for the vibration input; The vibration acceleration is the current and the past N sampling points; The autoregressive coefficients of historical resistance fluctuations; This is a constant bias term; in this embodiment, the model order is N=5 and M=2. The transfer function is used to estimate the resistance change component directly excited by mechanical vibration at the current moment. And it is labeled as passive fluctuation.

[0072] Subsequently, a signal separation operation is performed, subtracting the passive fluctuation calculated by the model from the measured instantaneous fluctuation of the contact resistance to obtain the residual sequence after removing external mechanical interference. For the above case, the calculated residual is 3 milliohms. The residual sequence is then subjected to stationarity testing (e.g., ADF unit root test) and trend analysis. If the residual sequence, after removing the vibration component, exhibits random white noise around zero, the contact surface is considered to be in good condition; if the residual sequence shows a non-negligible DC bias or a monotonically increasing trend (e.g., the residual value gradually drifts from 0.5 milliohms last month to the current 3 milliohms), it is determined that this is due to actual degradation of the contact surface caused by thickening of the oxide film, electrochemical corrosion, or arc erosion pits. These intrinsic change characteristics reflecting physical wear, electrochemical corrosion, and contact surface ablation of the contact material are extracted and labeled as active fluctuations.

[0073] This embodiment effectively improves the problem of false alarms in contact resistance test results caused by severe mechanical vibration of urban rail trains through cross-correlation analysis and transfer function fitting. It can quantify and isolate passive fluctuations caused by vibration, and extract active fluctuations reflecting the degradation of the physical and chemical properties of the contact material, thereby ensuring the accuracy of relay status monitoring in dynamic operating environments.

[0074] Furthermore, the specific implementation process of establishing a multi-factor health assessment model based on the aforementioned arc energy, active fluctuations, and passive fluctuations includes:

[0075] The system receives the arc energy, active fluctuations, and passive fluctuations. It then uses the relay's rated parameter set to normalize the arc energy and active fluctuations, generating electrical damage factor data and mechanical damage factor data. For the passive fluctuations, it applies nonlinear mapping logic to generate operating condition correction factor data reflecting the degree of external environmental interference. A multi-dimensional state feature space is constructed, mapping the electrical damage factor data, mechanical damage factor data, and operating condition correction factor data into state vectors within the feature space. These state vectors are then set as input variables to establish a multi-factor health assessment model.

[0076] Specifically, the system receives data on arc energy, active fluctuations, and passive fluctuations for a single operation or within a monitoring cycle. To eliminate differences between different physical units (such as the energy unit joule and the resistance unit ohm), the data is normalized using the relay's rated parameter set. Taking a certain type of subway train safety relay as an example, its technical specifications specify a cumulative arc energy threshold of 500 joules over the entire contact lifespan, and a maximum allowable contact resistance increment of 50 milliohms per operation. Assuming the currently measured single arc energy is 15 milliohms, it is compared with a single rated benchmark (such as 20 milliohms) or a lifespan conversion benchmark to generate electrical damage factor data with values ​​between 0 and 1 (e.g., 0.75, indicating a large arc impact). Simultaneously, the ratio of the measured active fluctuations caused by material degradation (e.g., 3 milliohms) to the failure threshold (50 milliohms) is calculated to generate mechanical damage factor data reflecting the degree of physical wear on the contact surface (e.g., 0.06).

[0077] Nonlinear mapping logic is applied to the passive fluctuations reflecting external vibration interference. Since the impact of mechanical vibration on relay life is not linear (i.e., slight vibrations have negligible effects, but severe vibrations exceeding a certain threshold exponentially accelerate aging), the S-shaped Sigmoid function is used as the mapping rule. When the separated passive fluctuations (virtual resistance increase caused by vibration) are less than 5 milliohms, the generated operating condition correction factor data approaches 1.0 (standard operating condition); when the passive fluctuations reach 20 milliohms, the operating condition correction factor generated by nonlinear mapping may rise sharply to 2.5, indicating that the vehicle is currently under extremely severe mechanical stress. To ensure the uniqueness and accuracy of the operating condition correction factor calculation, this embodiment explicitly defines the specific function model and key parameters of the aforementioned nonlinear mapping. When using the S-shaped Sigmoid function as the mapping rule, the improved mathematical expression is as follows:

[0078] ;

[0079] in For the separated passive fluctuations; As the baseline parameter, in this embodiment, we take... ; For the sensitivity coefficient, in this embodiment, we take... ; The inflection point threshold is taken in this embodiment. .

[0080] Based on this, a multi-dimensional state feature space is constructed, mapping the processed electrical damage factor data, mechanical damage factor data, and operating condition correction factor data to a state vector within this feature space. For example, a vector is constructed. A multi-factor health assessment model is established by setting the state vector as the input variable. The specific construction process of the multi-factor health assessment model is as follows: The rated arc energy of a single relay operation and the allowable increment of contact resistance are set as benchmark parameters. The measured arc energy value and the identified active fluctuation value are divided by the corresponding benchmark parameters to calculate the standardized electrical damage factor and mechanical damage factor. Since mechanical vibration has a nonlinear accelerating effect on relay aging, an S-shaped function is used as the mapping rule to calculate the separated passive fluctuation characteristics. This makes the correction coefficient for the operating condition approach the standard value of one under weak vibration, while the correction coefficient increases when severe vibration causes passive fluctuations. Therefore, the current operating condition type is determined based on the ratio of load current to rated current and the ratio of vibration acceleration.

[0081] This embodiment improves upon the limitation of relay monitoring relying solely on a single parameter (such as resistance or frequency) by establishing a multi-dimensional state feature space encompassing electrical damage, mechanical wear, and environmental conditions. By introducing a condition correction factor, it can identify the different meanings of the same resistance change under different vibration environments (a high resistance measured under severe vibration may be normal, while a high resistance measured under static conditions indicates a fault), and compensate for accelerated aging caused by environmental stress. This multi-factor fusion assessment model significantly improves the robustness of relay remaining life prediction, ensuring adaptability in the complex and variable operating environment of urban rail trains.

[0082] Furthermore, the specific implementation process of dynamically adjusting the weighting coefficients of arc energy and contact resistance in the health assessment based on the load current magnitude and vibration signal intensity during the current contact operation includes:

[0083] The relay load characteristic curve and vibration grading threshold table are called, and the load current magnitude is converted into a load stress index and the vibration signal intensity is converted into a mechanical interference index using a nonlinear mapping function. A two-dimensional weighted decision matrix based on fuzzy logic is constructed. The load stress index and mechanical interference index are used as index variables to query and calculate the proportion of damage contribution of arc energy and contact resistance under the current working condition, and load it into the parameter configuration port of the multi-factor health assessment model.

[0084] Specifically, the effective value of the load current at the moment of the operation (e.g., 8.5A) and the intensity of the environmental vibration signal within the time window before and after the operation (e.g., RMS value of 0.15g) are obtained. To quantify the impact of these physical quantities on the relay lifespan, the relay load characteristic curve and vibration classification threshold table pre-stored in the database are retrieved. The load characteristic curve describes the logarithmic decay relationship between the relay contact lifespan (number of operations) and the load current; the vibration classification threshold table sets the mechanical stress limits corresponding to different intensity levels according to the impact vibration test standard for rail transit on-board equipment (e.g., IEC 61373), and uses a nonlinear mapping function to perform dimensionless transformation on the above inputs. For the load current, an exponential mapping function is used to convert the current magnitude of 8.5A into a load stress exponent. For example, if the exponent corresponding to the rated current of 10A is 1.0, then the calculated load stress exponent for 8.5A is 0.72, indicating high electrical stress. For the vibration signal, the vibration intensity of 0.15g is mapped to a mechanical interference exponent. If the reference vibration threshold for inducing mechanical wear is set to 0.5g, then the mechanical interference index corresponding to the current vibration of 0.15g is only 0.1, indicating that the mechanical influence is small. The universe of discourse for both the normalized load stress index and the mechanical interference index is set to [0, 1], and three fuzzy subsets, "low", "medium", and "high", are defined for each input variable. For the load stress index, a combination of trapezoidal and triangular membership functions is used: the "low" subset uses a Z-shaped membership function, with a membership degree of 1 in the interval [0, 0.3], linearly decreasing to 0 at 0.5; the "medium" subset uses a triangular membership function, with a peak at 0.5 and a coverage range of [0.2, 0.8]; the "high" subset uses an S-shaped membership function, rising from 0.5 and saturating to 1 in the interval [0.7, 1.0]. For the mechanical interference index, considering the nonlinear destructive effect of vibration on contact reliability, its membership function distribution is adjusted to make it more sensitive to high-intensity vibration. The initial membership threshold of the "high" subset is shifted to the left to 0.4, meaning that when the mechanical interference index exceeds 0.4, the consideration of mechanical wear weight begins to increase significantly. The constructed two-dimensional weight decision matrix is ​​based on a complete fuzzy rule base, which contains nine core logics covering all working condition combinations. The domain of the output variable "arc energy weight coefficient" is set to [0, 1], divided into five fuzzy single-point values: "extremely low", "low", "medium", "high", and "extremely high". The specific fuzzy inference rules are as follows: when the load stress is "high" and the mechanical interference is "low", the output points to "extremely high" (0.9); when the load stress is "low" and the mechanical interference is "high", the output points to "extremely low" (0.1); when both the load stress and mechanical interference are "medium", the output points to "medium" (0.5).For complex operating conditions, such as when the load stress and mechanical disturbance are both "high," considering that severe vibration under high current will exacerbate the arc erosion effect, the rule sets arc damage to dominate, and the output is set to "high" (0.75). However, under light-load, stable operating conditions with "low" load stress and "low" mechanical disturbance, the rule sets the output to "medium" (0.5) to balance minor physical wear and electrical losses. Finally, the centroid method is used to defuzzify the fuzzy inference results, calculating precise weight values. This allows for continuous and smooth adjustment of the weight coefficients between 0.1 and 0.9, and the determined weight coefficients are then loaded into the parameter configuration port of the multi-factor health assessment model.

[0085] This embodiment introduces a two-dimensional weighted decision matrix based on fuzzy logic, which can identify the dominant damage mechanism at the current moment based on the actual current magnitude and vibration environment during each action, and dynamically adjust the weights of each factor in the model. This adaptive adjustment mechanism ensures that arc damage is prioritized under heavy load conditions and mechanical wear is prioritized under strong vibration conditions, thereby significantly improving the consistency between the health assessment results and the actual physical causes of failure, and greatly improving the accuracy of remaining life prediction.

[0086] Furthermore, the specific implementation process of accumulating the assessed single damage values ​​into the historical damage levels to monitor the relay's health status in real time and predict its remaining lifespan includes:

[0087] The historical damage data from the previous monitoring period is read and superimposed with the single damage value output by the multi-factor health assessment model to generate updated current cumulative damage data. The current cumulative damage data is compared with the relay failure threshold curve to calculate the proportion of the current damage level in the total lifespan capacity. Recent historical data segments are retrieved, and regression analysis algorithms are used to calculate the slope parameter of the damage growth trend. The current cumulative damage data is extrapolated to the future trend to calculate the estimated number of actions required to reach the failure threshold.

[0088] Specifically, the system reads historical damage data saved from the previous monitoring cycle. This data records the total standardized damage accumulated by the relay since it was put into use; assuming the historical cumulative value read is 25,000 standard damage units. Simultaneously, it acquires the single-transaction damage value, reflecting the overall loss of the current contact action, output by a multi-factor health assessment model (e.g., a single-transaction damage value calculated under strong arc and high vibration conditions is 1.5 standard damage units). These two values ​​are then superimposed to generate the updated current cumulative damage total data (i.e., 25,001.5 units), and this updated data is immediately written back to memory to overwrite the old data, completing the real-time status update.

[0089] The current cumulative damage data is compared with the relay failure threshold curve, which is the relay performance degradation limit trajectory measured by accelerated aging tests. To ensure the threshold curve has general engineering guidance significance, the specific construction and parameter calibration process of the accelerated aging test is as follows: A sample group of relays of the same model (e.g., 10 relays) is selected and placed in a temperature, humidity, and vibration composite stress test chamber. The test environment temperature is set to 85℃ to accelerate the chemical aging of the insulation and contact materials, while a random vibration spectrum conforming to IEC 61373 standard Class 1B is applied to simulate vehicle mechanical stress. Three key load levels are set for the test: light load (10% of rated current), rated load (100%), and overload (150% of rated current), and switching cycles are performed at the rated frequency. During the test, the contact resistance and arcing energy are recorded every 1000 operations. When the contact resistance exceeds 500 milliohms for 10 consecutive times or contact adhesion occurs, it is judged as physical failure. Based on the Weber distribution, the sample failure data is statistically analyzed to obtain the characteristic life (number of operations) under different load stresses. A power-law model was used to fit and establish the decay function relationship between load current and lifetime, i.e., the aforementioned relay load characteristic curve. Based on this, a single action damage under standard operating conditions (rated load, standard vibration) was defined as 1 standard unit. The total number of actions at physical failure was converted into a total standard damage amount (e.g., 100,000 units), and this total amount was used as a normalized failure threshold benchmark. Through the above standardized calibration process, individual differences were eliminated, generating a failure threshold curve for online comparison. The relay failure threshold curve defines the cumulative damage limit value (e.g., 100,000 standard damage units) corresponding to when the relay reaches the failure standard (e.g., contact resistance permanently exceeds 500 milliohms or engagement time delay exceeds 30 milliseconds). The proportion of the current cumulative damage amount (25001.5) in the total lifetime capacity (100,000) was calculated to obtain the current health status percentage (i.e., 25.0015% of the lifetime has been consumed), which was visually displayed as a progress bar on the host computer monitoring interface. The cumulative damage records of the most recent 1000 actions were extracted. Since the degradation process of relays often exhibits a non-linear accelerating characteristic (i.e., slow wear in the early stages and rapid wear in the later stages), regression analysis algorithms (such as least squares linear regression or exponential regression) are used to fit this data and calculate the slope parameter of the damage growth trend.

[0090] In the early stages of relay operation, the average damage value per action was 0.5 units. However, in the last 1000 actions, due to the near-complete wear of the contact surface plating, the average damage value per action increased to 1.2 units, resulting in a significantly larger calculated slope parameter, indicating an accelerated aging rate. Based on this latest slope parameter (1.2 units / action), the current cumulative damage data was extrapolated to predict future trends. The remaining lifespan capacity was calculated (100000 - 25001.5 = 74998.5 units) and divided by the current slope parameter to calculate the estimated number of actions required to reach the failure threshold (approximately 62499 actions). This predicted value was converted into remaining service life in days (assuming an average of 50 actions per day, approximately 1250 days remaining). When the predicted remaining lifespan falls below the set safety warning threshold, a replacement recommendation is automatically sent to the maintenance center.

[0091] This embodiment, through a prediction mechanism based on cumulative damage and trend extrapolation, not only records "how many actions occurred," but more importantly, it records "the quality and destructive force of each action," and captures the inflection point of accelerated relay performance degradation through regression analysis. This avoids waste caused by premature replacement and prevents malfunctions caused by undetected accelerated aging.

[0092] Example 2:

[0093] This embodiment provides an online status monitoring system for urban rail train relays, applicable to urban rail train relays, with reference to... Figure 2 The system includes a multi-dimensional data acquisition module, a feature fluctuation differentiation module, and a status health monitoring module.

[0094] The multi-dimensional data acquisition module uses an electrical sensor array installed on the urban rail train to collect relay coil current, contact voltage, load current, and environmental vibration signals; when the coil current undergoes a step change, it automatically locks and synchronously captures the electromechanical feature data set within the time window before and after the change.

[0095] The characteristic fluctuation differentiation module monitors the arcing phenomenon of contact voltage and load current during contact operation, performs time integration on the product of voltage and current during arcing, and calculates the arc energy; in the stable state of engagement, it calculates the contact resistance and fluctuation characteristics of the contact, and performs correlation analysis on the electromechanical characteristic dataset in conjunction with the environmental vibration signal to distinguish between passive fluctuations caused by vibration and active fluctuations caused by contact deterioration.

[0096] The health monitoring module establishes a multi-factor health assessment model based on the arc energy, active fluctuations, and passive fluctuations. It dynamically adjusts the weighting coefficients of arc energy and contact resistance in the health assessment according to the load current magnitude and vibration signal intensity at the time of contact action. It accumulates the assessed single damage value into the historical damage value, monitors the health status of the relay in real time, and predicts the remaining life.

[0097] Furthermore, the multi-dimensional data acquisition module is connected via shielded cables to an electrical sensor group installed near the relay and its base. This electrical sensor group consists of a closed-loop Hall current sensor (range 0-200mA) for monitoring the control loop, a high-impedance differential voltage probe (bandwidth DC-1MHz) for monitoring the voltage drop across the contacts, an open-loop Hall current sensor (range 0-50A) for monitoring the load loop, and a MEMS accelerometer (sensitivity 1000mV / g) fixed to the relay mounting plate. During operation, driven by a unified hardware clock, analog signals are acquired in parallel at a sampling rate of 100kHz, generating a continuous digital sequence stream and writing it in real-time into a high-speed ring buffer. This buffer is configured to cover at least 2 seconds of historical data depth to ensure that previous states can be retrieved when an event occurs. The data stream of the coil current channel is monitored in parallel, and sliding differential calculations are performed to monitor the rate of current change. Taking a certain model of DC110V train safety relay as an example, when the relay receives a closing command, and the coil current surges from 0mA to 45mA within 1 millisecond, it detects that the rate of change of current amplitude exceeds the trigger threshold (e.g., 20mA / ms), and immediately determines it as an event of "step change in coil current," locking this moment as event zero point T0. Once the triggering condition is met, the multi-dimensional data acquisition module immediately performs automatic locking and synchronous interception operations, and based on the event zero point T0, it backward intercepts 50 milliseconds of pre-trigger data (including background noise and initial vibration state before coil energization) and forward intercepts 450 milliseconds of post-trigger data (fully covering contact closure jump, arcing process, and steady-state closing process). All channel data within this 500-millisecond time window are stamped with a uniform nanosecond-level timestamp to eliminate transmission delay differences between channels. Finally, the aligned coil current, contact voltage, load current, and environmental vibration data are packaged and encapsulated to generate an electromechanical feature dataset containing transient and steady-state processes.

[0098] Furthermore, the feature fluctuation differentiation module receives the electromechanical feature dataset and extracts the digital sequences of contact voltage and load current. Using a first-order differential algorithm, the contact voltage sequence is scanned. When the voltage change rate exceeds a jump threshold (e.g., 10V / μs), it is determined to be the arc initiation moment; when the voltage amplitude subsequently drops to near zero volts (during engagement) or stabilizes at the power supply voltage (during disconnection) and the change rate returns to zero, it is determined to be the arc end moment. This locks in the arc time window (e.g., from 12.5 ms to 13.8 ms, lasting 1.3 ms). Within this window, the product of the contact voltage value and the load current value at each sampling moment is accumulated and summed to calculate the arc energy of a single operation. For example, under the condition of disconnecting a DC110V / 10A load, the calculated arc energy released is 14.5 millijoules, which is recorded as a key indicator for evaluating the degree of contact electrical erosion. In the stable engagement state, the contact voltage waveform is monitored. When the variance of the voltage amplitude is below the convergence threshold (e.g., 50mV) for 20 consecutive milliseconds, the relay is confirmed to have entered a mechanical steady state. The contact voltage sequence is divided by the load current sequence to generate an instantaneous contact resistance data stream. To eliminate electromagnetic noise, a statistical smoothing filtering algorithm (e.g., moving average filtering) is used to remove outliers and calculate the average contact resistance (e.g., 15 milliohms). Discreteness analysis is performed on the unsmoothed resistance data stream, and the standard deviation (e.g., 0.5 milliohms) is calculated as the initial fluctuation characteristic characterizing the stability of the contact resistance. Correlation analysis and signal decoupling are performed in conjunction with the environmental vibration signal to distinguish between passive fluctuations caused by vibration and active fluctuations caused by contact deterioration. The time-domain waveform of the environmental vibration at the same time is retrieved and strictly aligned with the time axis of the contact resistance data through polynomial interpolation. The correlation coefficient between the two in the time domain is calculated using a sliding window cross-correlation function. Assuming that a train travels to the wheel-rail joint and generates an impact vibration with an amplitude of 0.6g, the contact resistance is monitored to fluctuate instantaneously from 15 milliohms to 25 milliohms. Cross-correlation analysis revealed a correlation coefficient of 0.85 between the resistance fluctuation and the vibration waveform, indicating that the fluctuation was primarily caused by mechanical vibration. The transfer function was then fitted using the least squares method, estimating the theoretical resistance increase caused by the 0.6g vibration to be 9.2 milliohms, which was labeled as a passive fluctuation. Subtracting this passive fluctuation from the total fluctuation amplitude (10 milliohms) yielded a residual of 0.8 milliohms. After a smoothness test, this 0.8 milliohm residual was determined to be the actual resistance change caused by contact oxidation or wear, i.e., an active fluctuation.

[0099] Furthermore, the condition health monitoring module receives arc energy, active fluctuation, and passive fluctuation data, and performs multi-factor normalization mapping. It calls upon the built-in relay rated parameter set (including maximum withstand arc energy of 20mJ, failure contact resistance threshold of 50mΩ, etc.) to convert the measured physical quantities into dimensionless damage factors. For example, if the monitored arc energy is 18mJ, the active resistance fluctuation is 2mΩ, and the passive fluctuation (caused by vibration) is 5mΩ, the calculated electrical damage factor is 0.9 (close to the limit), and the mechanical damage factor is 0.04 (slight wear). For passive fluctuations, nonlinear mapping logic (such as the Sigmoid function) is applied to generate condition correction factor data reflecting the degree of external environmental interference (e.g., due to significant vibration, the correction factor is 1.2, indicating that a harsh environment will accelerate aging). Subsequently, a multi-dimensional state feature space is constructed, mapping the above factors to a state vector [0.9, 0.04, 1.2] within this space, which serves as the input variable for the subsequent evaluation model. Based on the load current magnitude (e.g., 9.5A) and vibration intensity (e.g., 0.1g) during the current action, a two-dimensional weighted decision matrix based on fuzzy logic is queried. When the load current is close to the rated value (10A), it is converted into a high load stress index; while when the vibration intensity is low, it is converted into a low mechanical disturbance index. According to the fuzzy rules, "electric arc ablation" is determined to be the dominant factor causing damage at this time. Therefore, the weight coefficients are dynamically adjusted, increasing the weight of arc energy to 0.85 and decreasing the weight of contact resistance to 0.15. The single damage value of this action is obtained by combining the state vector for weighted calculation. Conversely, under the condition of low current and strong vibration, the weight allocation may be reversed to arc weight 0.1 and resistance weight 0.9 to reflect the dominance of mechanical wear. The historical damage data stored in the previous monitoring cycle (e.g., accumulated to 45,000 units) is read, and the 0.77 units calculated in this case are added to generate the updated current accumulated total damage data (45,000.77). The module compares this total amount with the relay's life-cycle failure threshold curve (e.g., total capacity of 100,000 units), calculating that approximately 45% of the current lifespan has been consumed. To predict the remaining lifespan, the module retrieves recent historical data segments (e.g., the past 1,000 operations) and uses a linear regression algorithm to calculate the slope parameter of the damage growth trend. Recently, due to contact plating wear, the average damage value per operation has increased from 0.5 initially to 1.2 units / operation. Based on the latest slope parameter (1.2), the remaining capacity (54,999.23) is extrapolated, calculating that the estimated number of operations required to reach the failure threshold is approximately 45,832. Assuming an average of 50 operations per day for the train, the relay is predicted to reach the critical failure point in approximately 916 days, generating a corresponding early warning maintenance work order.

[0100] This embodiment incorporates a multi-dimensional data acquisition module. Through parallel and synchronous acquisition, it ensures the alignment of mechanical vibration signals and electrical performance signals on the time axis, providing a high-fidelity data foundation for accurately distinguishing between passive fluctuations caused by external vibrations and active fluctuations caused by internal contact degradation. The design of automatically locking and capturing time windows before and after changes ensures complete capture of key transient features, including arcing and bouncing, while avoiding the generation of massive amounts of invalid data during non-operational periods, significantly reducing the storage pressure and transmission bandwidth requirements for relay monitoring. A feature fluctuation differentiation module, through time-domain segmentation processing and decoupling from multi-source signals, can quantify and isolate false faults (passive fluctuations) caused by environmental vibrations, and identify true faults (active fluctuations) caused by the physical degradation of contact materials. This intelligent feature separation mechanism provides reliable data input for subsequent health assessment and lifespan prediction. A condition health monitoring module, through the establishment of a dynamic weighted multi-factor evaluation model, can adaptively adjust the evaluation strategy according to the specific environment (current and vibration) of each operation, accurately quantifying the actual wear and tear on the relay's lifespan caused by each operation. By combining a trend extrapolation algorithm based on regression analysis, it is possible to capture the inflection point when relay performance enters the accelerated degradation period, thereby providing a more accurate prediction of remaining life.

[0101] Example 3:

[0102] This embodiment fully deploys the above-described method and system for online monitoring of relay status in urban rail transit in a type B train of a certain urban rail transit system, referring to... Figure 3 This enables online monitoring of the status of relays on urban rail trains.

[0103] Furthermore, a high-sensitivity MEMS accelerometer is attached to the back of the relay mounting base, and Hall current sensors are inserted into the relay coil circuit and the contact load circuit, respectively. A high-impedance voltage acquisition probe is connected in parallel across the contacts. During train operation, the door controller issues an open or close command, energizing the relay coil circuit. When the coil current is detected to jump from 0 mA to the rated operating current at a rate exceeding 50 mA per millisecond, the system determines a step change has occurred, immediately locks this moment as the zero point of action, and automatically retrieves coil current, contact voltage, load current, and environmental vibration data within a time window from 50 milliseconds before the action to 450 milliseconds after the action using FPGA hardware synchronization technology. All channel data undergoes time-series alignment processing, eliminating microsecond-level errors caused by sensor transmission delays and generating a standardized electromechanical feature dataset.

[0104] Furthermore, during the transient period of contact operation, the system extracts the contact voltage data sequence and identifies the interval of drastic voltage amplitude jumps by calculating the first derivative, thus pinpointing the arcing start and end times. For example, in a case of interrupting a 10-ampere inductive load, the system locks the arcing duration at 1.5 milliseconds. The system numerically integrates the voltage and current product within this time period, calculating the arc energy of a single operation to be 18.5 millijoules, which directly reflects the degree of physical ablation of the contact material due to the high-temperature arc. During the steady-state period after the relay engages, the system performs in-depth analysis of the contact resistance. At this time, the train is passing through a section of heavily worn track, and the root mean square value of vibration acceleration collected by the environmental vibration sensor is as high as 0.4g. The system determines that the relay has ended its mechanical bounce and entered the conducting state by voltage waveform convergence, and then calculates the voltage-to-current ratio to obtain the instantaneous contact resistance sequence. The raw data shows that the contact resistance fluctuates drastically between 20 milliohms and 35 milliohms. To identify whether this is a relay fault, the system performs correlation analysis of the environmental vibration signal. Through sliding window cross-correlation calculations, a strong correlation coefficient of 0.88 was found between the resistance fluctuation waveform and the environmental vibration waveform in the time domain. The system used the least squares method to fit the transfer function, estimating the "passive fluctuation" component directly excited by the 0.4g vibration to be approximately 12 milliohms. Subtracting this passive fluctuation from the original measurement value yielded the residual sequence after removing mechanical interference. After stationarity testing, this residual stabilized at around 23 milliohms. This value represents the active fluctuation caused by contact deterioration, reflecting the thickening of the oxide film on the contact surface and the microscopic wear of the contact surface.

[0105] Furthermore, the system normalizes the calculated 18.5 millijoules of arc energy and 23 milliohms of active contact resistance fluctuation to generate electrical and mechanical damage factors. Combining the load current magnitude (10 amps, heavy load) and vibration intensity (0.4g, moderate vibration) during the current operation, it queries the built-in two-dimensional weighted decision matrix based on fuzzy logic. Since the load current is relatively large at this time, the fuzzy logic determines that electrical ablation is the main failure cause, therefore dynamically adjusting the weight coefficients, assigning a weight of 0.75 to the arc energy and 0.25 to the contact resistance. Based on this, the system calculates the comprehensive single-instance damage value for this operation. Historical cumulative damage data from the previous monitoring cycle is read and overlaid with the calculated single-instance damage value for updating. The system compares the current cumulative damage amount with the full life-cycle failure threshold curve, determining the current health level to be 82%. To predict the remaining lifespan, the system calls upon historical data from the last 500 operations and uses a regression analysis algorithm to calculate the slope of the damage growth trend. Analysis results showed an accelerating rate of damage growth. Based on this, the system extrapolated future trends and calculated that the relay was expected to reach its failure threshold after approximately 35,000 operations. According to the train's average daily door opening and closing frequency, the system predicted a remaining service life of 650 days. This prediction, along with a warning message recommending monitoring during secondary maintenance, was sent to the operations and maintenance center, thus enabling online monitoring of the relay's status on the urban rail train.

[0106] This embodiment effectively improves the problem of inaccurate contact resistance testing and high false alarm rate caused by severe vibration in the rail transit vehicle environment by synchronously acquiring multi-dimensional data and decoupling vibration, and achieves accurate differentiation between passive interference and active faults. By introducing a dynamic weight adjustment mechanism for load current and vibration intensity, the health assessment model can adapt to different operating conditions, and can accurately capture both high-current arc erosion and low-current mechanical wear. Based on trend extrapolation prediction of historical data, the operation and maintenance efficiency and safety of urban rail trains are greatly improved.

[0107] It should be clarified that the embodiments described above are merely exemplary and are intended to aid in understanding the present invention, not to limit it. Those skilled in the art can make various changes and modifications after grasping the core ideas of the present invention. Therefore, the scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for online monitoring of the status of relays in urban rail transit trains, characterized in that, include: Using an electrical sensor array installed on an urban rail train, the coil current, contact voltage, load current, and environmental vibration signals of the relay are collected; when the coil current undergoes a step change, the electromechanical characteristic data set within the time window before and after the change is automatically locked and synchronously captured. During contact operation, the arcing phenomenon of contact voltage and load current is monitored, and the product of voltage and current during the arcing period is integrated over time to calculate the arc energy; under the stable engagement state, the contact resistance and fluctuation characteristics of the contact are calculated, and the correlation analysis of the electromechanical characteristic dataset is performed in conjunction with the environmental vibration signal to distinguish between passive fluctuations caused by vibration and active fluctuations caused by contact deterioration. A multi-factor health assessment model is established based on the arc energy, active fluctuations, and passive fluctuations. The weighting coefficients of arc energy and contact resistance in the health assessment are dynamically adjusted according to the load current and vibration signal intensity at the time of contact action. The assessed single damage value is accumulated into the historical damage value to monitor the health status of the relay in real time and predict the remaining life.

2. The method for online monitoring of the status of urban rail train relays according to claim 1, characterized in that, The specific process of collecting relay coil current, contact voltage, load current, and environmental vibration signals using an electrical sensor array installed on an urban rail train includes: An electrical sensor group is configured on the relay control circuit to continuously sense the magnetic field changes generated by the excitation of the coil and generate a load current signal; the load current signal reflecting the state of the controlled equipment is captured in the execution circuit to generate a digital sequence vector of coil current, contact voltage, load current and environmental vibration with timestamp.

3. The method for online monitoring of the status of urban rail train relays according to claim 1, characterized in that, The specific implementation process of automatically locking and synchronously capturing the electromechanical characteristic data set within the time window before and after the change when the coil current undergoes a step change includes: A sliding difference operation is performed on the digital sequence vector to calculate the amplitude change rate between adjacent sampling points, and the amplitude change rate is logically compared with the action trigger threshold. When the amplitude change rate exceeds the action trigger threshold, it is determined that a step change has occurred and a time anchor signal is generated. The time anchor signal is time-aligned and the transmission delay difference is eliminated. The aligned multidimensional time series is encapsulated into an electromechanical feature dataset containing transient and steady-state processes.

4. The method for online monitoring of the status of urban rail train relays according to claim 1, characterized in that, During contact operation, the arcing phenomenon of contact voltage and load current is monitored. The product of voltage and current during arcing is integrated over time to calculate the specific realization process of arc energy, which includes: The contact voltage sequence and load current sequence are extracted from the electromechanical feature dataset. The first-order differential operation is performed on the contact voltage sequence to identify the transient interval where the voltage amplitude changes drastically and to locate the time range of contact action. The numerical integration algorithm is applied to accumulate and sum the voltage and current of the arc phenomenon within the time range of contact action to calculate the arc energy characterizing the degree of contact ablation.

5. The method for online monitoring of the status of urban rail train relays according to claim 1, characterized in that, The specific implementation process for calculating the contact resistance and fluctuation characteristics of the contacts under the stable engagement state includes: The starting point of the steady-state engagement is determined by the amplitude convergence of the contact voltage waveform. The contact voltage sequence and load current sequence after the starting point are extracted from the electromechanical feature dataset, and point-by-point synchronous division is performed to generate an instantaneous contact resistance data stream containing time-varying characteristics. A statistical smoothing filtering algorithm is used to remove outlier noise points in the instantaneous contact resistance data stream to obtain the contact resistance. At the same time, the discreteness analysis of the instantaneous contact resistance data stream is performed to calculate the variance and standard deviation statistics and extract the fluctuation characteristics that characterize the stability of the contact resistance.

6. The method for online monitoring of the status of urban rail train relays according to claim 1, characterized in that, The specific process of performing correlation analysis on the electromechanical feature dataset based on the environmental vibration signals to distinguish between passive fluctuations caused by vibration and active fluctuations caused by contact deterioration includes: The environmental vibration time-domain waveform and instantaneous fluctuation sequence of contact resistance are retrieved from the electromechanical feature dataset. Frequency normalization is performed using a polynomial interpolation algorithm to establish a time-axis aligned synchronous analysis vector. The correlation coefficient flow in the time domain is calculated using a sliding window cross-correlation function to identify and lock the time segments exhibiting correlation. The transfer function is established by fitting using the least squares method to estimate the resistance change component directly excited by mechanical vibration and label it as passive fluctuation. The passive fluctuation is subtracted from the instantaneous fluctuation of contact resistance to obtain the residual sequence after removing external mechanical interference. The stationarity test and trend analysis are performed on the residual sequence to extract the intrinsic change characteristics reflecting the physical wear, electrochemical corrosion, and contact surface ablation of the contact material and label them as active fluctuation.

7. The method for online monitoring of the status of urban rail train relays according to claim 1, characterized in that, The specific implementation process of establishing a multi-factor health assessment model based on the aforementioned arc energy, active fluctuations, and passive fluctuations includes: The system receives the arc energy, active fluctuations, and passive fluctuations. It then uses the relay's rated parameter set to normalize the arc energy and active fluctuations, generating electrical damage factor data and mechanical damage factor data. For the passive fluctuations, it applies nonlinear mapping logic to generate operating condition correction factor data reflecting the degree of external environmental interference. A multi-dimensional state feature space is constructed, mapping the electrical damage factor data, mechanical damage factor data, and operating condition correction factor data into state vectors within the feature space. These state vectors are then set as input variables to establish a multi-factor health assessment model.

8. The method for online monitoring of the status of urban rail train relays according to claim 1, characterized in that, The specific implementation process of dynamically adjusting the weighting coefficients of arc energy and contact resistance in health assessment based on the load current magnitude and vibration signal intensity during each contact operation includes: The relay load characteristic curve and vibration grading threshold table are called, and the load current magnitude is converted into a load stress index and the vibration signal intensity is converted into a mechanical interference index using a nonlinear mapping function. A two-dimensional weighted decision matrix based on fuzzy logic is constructed. The load stress index and mechanical interference index are used as index variables to query and calculate the proportion of damage contribution of arc energy and contact resistance under the current working condition, and load it into the parameter configuration port of the multi-factor health assessment model.

9. The method for online monitoring of the status of urban rail train relays according to claim 1, characterized in that, The specific process of accumulating the assessed single damage values ​​into the historical damage levels to monitor the relay's health status in real time and predict its remaining life includes: The historical damage data from the previous monitoring period is read and superimposed with the single damage value output by the multi-factor health assessment model to generate updated current cumulative damage data. The current cumulative damage data is compared with the relay failure threshold curve to calculate the proportion of the current damage level in the total lifespan capacity. Recent historical data segments are retrieved, and regression analysis algorithms are used to calculate the slope parameter of the damage growth trend. The current cumulative damage data is extrapolated to the future trend to calculate the estimated number of actions required to reach the failure threshold.

10. An online monitoring system for the status of relays in urban rail transit trains, characterized in that, include: The multi-dimensional data acquisition module uses an electrical sensor array installed on the urban rail train to collect relay coil current, contact voltage, load current, and environmental vibration signals; when the coil current undergoes a step change, it automatically locks and synchronously captures the electromechanical feature data set within the time window before and after the change. The characteristic fluctuation differentiation module monitors the arcing phenomenon of contact voltage and load current during contact operation, performs time integration on the product of voltage and current during arcing, and calculates the arc energy; in the stable state of engagement, it calculates the contact resistance and fluctuation characteristics of the contact, and performs correlation analysis on the electromechanical characteristic dataset in conjunction with the environmental vibration signal to distinguish between passive fluctuations caused by vibration and active fluctuations caused by contact deterioration. The health monitoring module establishes a multi-factor health assessment model based on the arc energy, active fluctuations, and passive fluctuations. It dynamically adjusts the weighting coefficients of arc energy and contact resistance in the health assessment according to the load current magnitude and vibration signal intensity at the time of contact action. It accumulates the assessed single damage value into the historical damage value, monitors the health status of the relay in real time, and predicts the remaining life.

Images