Method for testing cycle life of marine charging power supply

Abstract

The invention discloses a method for testing the cycle life of a marine charging power supply, particularly relates to the technical field of power supply equipment life testing, and is used for solving the problem of passive maintenance caused by difficulty in early recognition of continuous degradation and prediction of residual available processes due to the fact that an existing accelerated life test focuses on failure judgment. The method comprises the following steps of: driving a start-stop and load mutation task sequence by taking the shell temperature of a key power device as a cyclic trigger condition, generating a health characteristic value of a relative reference at a node switching position, serializing and recording, and calculating a shell temperature over-reference heat occupancy index and a ripple spectrum side band fingerprint index for adjacent node windows; and inputting a Gaussian mixture model to obtain a degradation credible coefficient, judging a degradation section boundary, extrapolating a residual available process based on a degradation section change speed, and exporting a complete degradation track data packet, thereby realizing reliable life prediction and health management support when the equipment is not failed.

Description

Technical Field

[0001] This invention relates to the field of power supply life testing technology, and more specifically, to a method for testing the cycle life of marine charging power supplies. Background Technology

[0002] Marine charging power supplies typically need to maintain the stability of power conversion and charging control under conditions of long-term continuous operation, frequent start-stop cycles, and load fluctuations. Once anomalies occur, timely replacement and repair are often difficult. Maintenance personnel prefer to identify, before failures occur, whether equipment is slowly degrading, to what extent it has degraded, and how long it can remain stably operational. This allows maintenance to shift from reactive emergency repairs to planned health management and lifespan prediction. To achieve this goal, cyclic life testing not only verifies operational functionality but also continuously accumulates state information during the cycle that can be used for health assessment and remaining lifespan estimation. This is because failure prediction and health management inherently emphasize the identification of early damage and system degradation, as well as the reliable prediction of remaining lifespan.

[0003] In the prior art, for example, the Chinese invention patent application "A Method and Apparatus for Accelerated Life Testing of Charging Equipment" (Publication No. CN109212340A) gives a typical accelerated life testing approach: first, the conventional test time is obtained based on factors such as the design life, then temperature is selected as the accelerated stress to calculate the acceleration factor, and then a cyclic test is carried out according to the set temperature change curve until the accelerated life test time is reached, and failure is used as the main judgment; if the accelerated life test is completed and failure is not judged, the failure is checked again by the load programmable switching and switching cycle, and if necessary, the functions such as plugging and unplugging the charging gun are also verified. The starting point of this type of approach is to obtain life conclusions as quickly as possible and improve testing efficiency. Therefore, the testing process naturally revolves around point-to-point judgment and post-event review. The cyclical process is more like providing a basis for the final pass or fail than serving to model the degradation pattern. In actual operation, many faults do not jump to failure instantly, but first go through a gradual stage such as performance margin shrinkage, protection boundary approach, or intermittent anomalies. However, when the test output mainly stays at the end of the failure judgment, and the process data lacks continuous collection and staged expression of health indicators, it is difficult to form a transferable degradation trajectory and labeling system. Even if fault prediction and health management are introduced later, it is often only possible to obtain post-event explanations or near-failure warnings. It is difficult to provide usable remaining life judgment and maintenance window in advance, resulting in the fact that although life tests have been carried out, they still cannot support truly implementable predictive maintenance.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for testing the cycle life of marine charging power supplies. This method uses the case temperature of key power devices as a cycle trigger condition to drive start-stop and load mutation task sequences. At node switching points, relative baseline health characteristic values ​​are generated and serialized. The case temperature exceedance thermal occupancy index and ripple spectrum sideband fingerprint index are calculated for adjacent node windows. These are input into a Gaussian mixture model to obtain degradation confidence coefficients and determine degradation segment boundaries. Based on the degradation segment change rate, the remaining usable history is extrapolated, and a complete degradation trajectory data package is exported. This enables reliable life prediction and health management support before equipment failure, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: S1: Connect the marine charging power supply under test to the standard input and controllable load for pre-running. After the case temperature of the key power device stabilizes, collect the case temperature and output ripple characteristic values ​​and set them as the reference case temperature and reference ripple, respectively. S2: Set the cycle triggering conditions with the reference case temperature, construct the task sequence of start-stop and load change, and drive node switching across the triggering conditions with the case temperature of key power devices to form a cycle driven by internal thermal state. S3: Synchronously collect the case temperature and output ripple characteristic values ​​of key power devices at each node switching point, generate health characteristic values ​​relative to the reference case temperature and reference ripple according to a unified offset rule, and store them as serialized records in sequence with the node order. S4: Extract thermal stress occupancy features and ripple spectrum morphology features from adjacent node windows, input the two types of features into a pre-trained model trained by healthy segment samples to obtain degradation confidence coefficients, and determine the degradation segment boundary and generate degradation segment identifiers when the degradation confidence coefficients continuously exceed the standard and the healthy feature values ​​continue to shift. S5: Based on the continuous history corresponding to the degradation segment identifier and the rate of change of health characteristic values, calculate the remaining usable history and output the lifetime conclusion. At the same time, export the degradation trajectory data package containing shell temperature, ripple, health characteristic values ​​and degradation confidence coefficient for health management.

[0007] Furthermore, step S1 includes: Connect the marine charging power supply under test to the standard input power supply and the controllable electronic load, and start the pre-run with a constant load; monitor the case temperature of key power devices in real time, calculate the absolute difference of case temperature between adjacent sampling times, and determine that it has entered the stable range after it is continuously less than the preset condition; select a continuous acquisition window from the starting point of the stable range, and continuously record the case temperature of key power devices and the peak-to-peak value of output ripple; weight and accumulate the case temperature of key power devices in the window according to the sampling time interval and divide it by the duration of the window to obtain the reference case temperature, and determine the maximum value of the peak-to-peak value of output ripple in the window as the reference ripple.

[0008] Furthermore, step S2 includes: The high-temperature trigger threshold is determined by adding a preset temperature rise offset to the reference case temperature, and the low-temperature trigger threshold is determined by subtracting a preset temperature drop offset from the reference case temperature. The charging power supply start-up and shutdown and load change are alternately organized into a task sequence of high-stress tasks and low-stress tasks.

[0009] Furthermore, step S2 also includes: The task sequence node switching is triggered when the current case temperature of the critical power device is equal to or exceeds the high temperature trigger threshold, switching from a high stress task to a low stress task; and when the current case temperature of the critical power device is equal to or below the low temperature trigger threshold, switching from a low stress task to a high stress task.

[0010] Furthermore, step S3 includes: At the moment of triggering node switching and before the switching action is executed, the case temperature and output ripple characteristic values ​​of key power devices are collected synchronously. The healthy characteristic case temperature is generated by subtracting the reference case temperature from the current case temperature and dividing by the reference case temperature. The healthy characteristic ripple is generated by subtracting the reference ripple from the current output ripple characteristic value and dividing by the reference ripple. The data are stored sequentially as serialized records along with the node.

[0011] Furthermore, step S4 includes: The over-temperature accumulation is obtained by integrating the portion of the critical power device case temperature that is higher than the reference case temperature within the adjacent node window of the serialized record over time. Then, the over-temperature accumulation is divided by the product of the window duration and the reference case temperature to obtain the case temperature over-reference heat occupancy index.

[0012] Furthermore, step S4 also includes: The short-time output ripple waveform within the adjacent node window is decomposed into a spectrum to extract the sideband spectral energy near the switching fundamental frequency. The sideband fingerprint index of the ripple spectrum is obtained by dividing it by the total ripple spectral energy. The shell temperature over-reference heat occupancy index and the sideband fingerprint index of the ripple spectrum are input into the Gaussian mixture model trained on the healthy segment samples to obtain the degradation confidence coefficient.

[0013] Furthermore, step S4 also includes: When the degradation confidence coefficient of multiple consecutive adjacent node windows continuously exceeds the preset high-order threshold and the health feature value continues to shift, the degradation segment boundary is determined to be established and a degradation segment identifier is generated.

[0014] Furthermore, step S5 includes: Based on the degraded segment identifier, extract the node subsequence of the degraded segment, calculate the difference in shell temperature and ripple of the healthy feature between the end and the beginning of the degraded segment, and divide them by the number of nodes corresponding to the degraded segment to obtain the rate of change. Take the larger of the two as the comprehensive rate of change.

[0015] Furthermore, step S5 also includes: The remaining usable history is obtained by subtracting the maximum value of the health characteristics at the end of the degradation segment from the preset failure threshold and dividing by the comprehensive change rate. The test conclusion is output, which includes the total number of nodes, the location of the degradation segment, the comprehensive change rate and the remaining usable history. The degradation trajectory data package containing the corresponding parameters of all nodes and the degradation segment identifier is exported.

[0016] The technical effects and advantages of the method for testing the cycle life of marine charging power supplies according to the present invention are as follows: This invention uses the case temperature of key power devices as the trigger for cyclic propagation, transforming the life testing of marine charging power supplies under typical operating conditions such as start-up, shutdown, and sudden load changes from an externally timed process to a cyclic process that unfolds naturally according to the internal stress state. This makes the testing process closer to the operating rhythm that is prone to degradation in actual use. At the same time, only two key information types, case temperature and output ripple, are collected at the cycle nodes, and a stable benchmark is formed on this basis. Furthermore, two types of indices reflecting thermal stress occupancy and ripple morphology changes are extracted, and degradation confidence coefficients are obtained using machine learning to participate in the determination of degradation segment boundaries. This transforms the degradation identification process, which is originally susceptible to interference from occasional fluctuations, into a decision-making process with consistent constraints. This reduces the possibility of misjudging short-term disturbances as life degradation and enables earlier and more reliable identification of continuous degradation and completion of degradation stage labeling before the equipment fails. As a result, the output life conclusion is no longer limited to the final judgment of whether it has failed, but simultaneously generates degradation trajectory data that can be directly used for fault prediction and health management. This allows the maintenance decision of marine charging power supplies to shift from ex-post response to state-based pre-planning, improving the usability of test conclusions and their engineering implementation value. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for testing the cycle life of a marine charging power supply according to the present invention. Detailed Implementation

[0018] 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.

[0019] Example 1: Figure 1 This invention provides a method for testing the cycle life of marine charging power supplies, comprising: S1: Connect the marine charging power supply under test to the standard input and controllable load for pre-running. After the case temperature of the key power device stabilizes, collect the case temperature and output ripple characteristic values, and set them as the reference case temperature and reference ripple, respectively.

[0020] S2: Set the cycle triggering conditions based on the reference case temperature, construct the task sequence of start-stop and load change, and drive node switching across the triggering conditions with the case temperature of key power devices to form a cycle driven by internal thermal state.

[0021] S3: Synchronously collect the case temperature and output ripple characteristic values ​​of key power devices at each node switching point, generate health characteristic values ​​relative to the reference case temperature and reference ripple according to a unified offset rule, and store them as serialized records in sequence with the node sequence.

[0022] S4: Extract thermal stress occupancy features and ripple spectrum morphology features from adjacent node windows. Input the two types of features into a pre-trained model trained by healthy segment samples to obtain degradation confidence coefficients. When the degradation confidence coefficients continuously exceed the standard and the healthy feature values ​​continue to shift, determine the degradation segment boundary and generate a degradation segment identifier.

[0023] S5: Based on the continuous history corresponding to the degradation segment identifier and the rate of change of health characteristic values, calculate the remaining usable history and output the lifetime conclusion. At the same time, export the degradation trajectory data package containing shell temperature, ripple, health characteristic values ​​and degradation confidence coefficient for health management.

[0024] This invention addresses the cyclic life testing requirements of marine charging power supplies under conditions of long-term continuous operation, frequent start-stop, and load fluctuations. It aims to shift from traditional end-of-life failure judgment to the construction of degradation trajectories that support fault prediction and health management. To achieve this goal, the entire testing scheme uses the internal thermal state of the equipment as the core driver of the cycle progression and continuously collects and processes key state information to identify early signs of degradation. Step S1, as the starting point of the entire testing process, focuses on establishing reliable health benchmark reference values—benchmark shell temperature and benchmark ripple—under controlled pre-operation conditions. These benchmark values ​​are derived from the stable operation phase of the equipment in a brand-new or healthy state and directly serve as a unified reference system for calculating health characteristic value offsets, normalizing thermal occupancy index, and determining degradation during subsequent cycles. Without a stable and comparable benchmark, the calculation of all subsequent relative offsets, health characteristic values, and degradation confidence coefficients will lack a consistent basis, leading to decreased sensitivity in degradation identification or increased risk of misjudgment. Therefore, step S1 ensures the representativeness and transferability of the baseline values ​​through standardized pre-running and strict stability interval determination, providing a solid data starting point for subsequent cyclic processes driven by internal thermal states and degradation segment labeling.

[0025] The specific implementation method of step S1 is as follows: S101: Connect the marine charging power supply under test to the standard input and controllable load and start pre-run. To ensure that subsequent benchmark values ​​reflect the inherent characteristics of the equipment under controlled and stable operating conditions, it is first necessary to eliminate the interference of external power fluctuations and load uncertainties on the internal state. Therefore, the equipment is operated in a closed test environment consisting of a standard input power supply and a programmable load. During this stage, no start-stop or abrupt changes are introduced; only constant operating conditions are maintained to allow critical power devices to gradually enter a normal thermal equilibrium state.

[0026] The specific process is as follows: Connect the input terminal of the marine charging power supply under test to a stable standard input power supply whose output voltage and frequency meet the rated requirements. Simultaneously, connect the output terminal to a controllable electronic load and set the electronic load to a constant load at the intermediate level of the rated power. Then, start the charging power supply to enter continuous operation mode and begin the pre-running process. At this time, the key power devices gradually accumulate heat during constant power conversion until the shell temperature tends to stabilize.

[0027] In one embodiment, on the test bench, the technician first connects the charging power input terminal to the rated 380V three-phase AC power supply provided by the laboratory, then connects the output terminal to the electronic load device and sets the load to a constant power mode of 60% of the rated power through the panel. Then, the power start button is pressed, and the device begins to continuously supply power to the load. The panel displays that the output voltage and current are stable near the rated values.

[0028] S102: Real-time monitoring of the case temperature of key power devices and determination of whether they have entered a stable range.

[0029] The representativeness of the benchmark value depends on whether the device has moved out of the transient temperature rise stage of initial power-on when the data is acquired. Therefore, it is necessary to continuously monitor the change trend of the case temperature of key power devices during the pre-operation process. Only when the case temperature change is extremely small can it be considered to have entered the stable range, thus providing a reliable time window for subsequent acquisition.

[0030] The specific processing procedure is as follows: A high-precision temperature sensor fixed to the surface of the critical power device continuously collects case temperature values. At fixed sampling intervals, the absolute difference between the current case temperature and the case temperature at the previous sampling time is calculated. When this absolute difference remains below a preset case temperature stability threshold for multiple consecutive sampling periods, and this state continues for a certain number of periods, it is determined that the case temperature of the critical power device has entered a stable range, and the start time of this stable range is recorded. Subsequently, the process continues for an additional period to confirm the sustainability of the stable state.

[0031] In one embodiment, the sensor collects the shell temperature every 5 seconds, and the system automatically calculates the absolute difference between two adjacent samples. When it is found that the difference is less than 0.2℃ for each of the 60 consecutive sampling cycles (approximately 5 minutes), and the same condition is still met after running for another 30 cycles, the test software marks the current moment as the starting point of the stable interval, and the screen displays a message indicating that the shell temperature has stabilized.

[0032] S103: Synchronously acquire data within the stable range and determine the reference shell temperature and reference ripple.

[0033] After confirming the case temperature stability of critical power devices, case temperature and output ripple data must be collected synchronously within the same time period to ensure that both correspond to the same healthy operating state. The reference case temperature is obtained by integrating the case temperature curve within the stable range over time and dividing by the range duration, which can accurately characterize the typical case temperature level under thermal equilibrium conditions. The reference ripple is taken as the maximum value of the peak-to-peak value of the output ripple within the stable range, which conservatively characterizes the most unfavorable ripple level that may occur under healthy conditions, thus providing sufficient sensitivity for subsequent degradation offset calculations.

[0034] The specific processing procedure is as follows: Starting from the beginning of the stable interval, a continuous acquisition window is selected. Within this window, all values ​​of the case temperature change of the key power device over time are continuously recorded, and the peak-to-peak sequence of the output voltage ripple is recorded simultaneously.

[0035] The calculation process for determining the reference shell temperature is as follows: First, the shell temperature values ​​at each sampling moment within the acquisition window are accumulated to obtain the total accumulated shell temperature. Then, the total accumulated temperature is divided by the duration of the acquisition window, and the quotient is the reference shell temperature. This process is equivalent to performing time integration on the shell temperature curve and then normalizing it, ensuring that the result is not affected by the sampling density and accurately reflects the average level of the shell temperature within the window.

[0036] The calculation process for determining the baseline ripple is as follows: Traverse all peak-to-peak values ​​of the output ripple within the acquisition window, find the peak-to-peak value with the largest value, and directly use it as the baseline ripple. This process can capture the highest fluctuation of the ripple under healthy conditions, ensuring that subsequent calculations of health characteristic value offsets have sufficient conservative margins and avoiding underestimation of early degradation signs.

[0037] In one embodiment, after confirming the stable range, the test system automatically opens a 30-minute acquisition window. During this period, the temperature sensor continuously transmits case temperature data, and the differential probe synchronously captures the output voltage ripple waveform and calculates the peak value each time. At the end of the window, the software first adds up all the case temperature samples within 30 minutes to obtain a sum, and then divides it by the total number of seconds in 30 minutes to obtain a reference case temperature of approximately 85°C; simultaneously, it selects the largest peak value of approximately 120mV from all ripple peak values ​​and records it directly as the reference ripple. Subsequently, the system saves these two reference values ​​and indicates that a healthy baseline has been established, allowing the system to enter the cyclic testing phase.

[0038] Step S1 involves connecting the marine charging power supply under test to a standard input power supply and a controllable electronic load to initiate pre-running under constant load, establishing a controlled healthy operating environment. Subsequently, the case temperature of key power devices is monitored in real time, and the absolute difference in case temperature between adjacent sampling times is calculated. Once a preset threshold is continuously met, the system is considered to have entered a stable range. A continuous sampling window is selected from the starting point of the stable range, continuously recording the case temperature of key power devices and the peak-to-peak value of output ripple. The case temperatures of key power devices within the window are weighted and accumulated according to the sampling time interval and divided by the window duration to obtain the reference case temperature. The maximum peak-to-peak value of the output ripple within the window is determined as the reference ripple, thus providing a unified and comparable health reference benchmark for subsequent cyclic triggering, health characteristic value generation, and degradation assessment.

[0039] This invention transforms the cycle life testing of marine charging power supplies from traditional external timing-driven methods to a natural cyclic process dominated by the internal thermal state of the equipment. This more realistically reproduces the thermal stress accumulation and degradation caused by start-up, shutdown, and load fluctuations during actual operation. Step S1 establishes a reference case temperature as a healthy thermal balance reference. Step S2 uses this reference to define high-temperature and low-temperature trigger thresholds, alternating start-up, shutdown, and load abrupt changes into a task sequence. Node switching is entirely triggered by the case temperature of key power devices crossing thresholds, forming an adaptive closed-loop cycle. This mechanism ensures that the cycle rhythm adjusts with the individual thermal response of the equipment, avoiding stress deviations caused by fixed timing sequences. This makes subsequent degradation data closer to the actual degradation rhythm, providing a reliable trajectory for early identification of continuous degradation.

[0040] The specific implementation method of step S2 is as follows: S201: Define the cycle triggering condition based on the reference shell temperature.

[0041] The reference shell temperature represents the thermal equilibrium level of the equipment under healthy conditions. In order for the cycle process to naturally unfold the heating and cooling phases around this level, it is necessary to set clear shell temperature boundary thresholds. These thresholds directly determine the timing of task switching, thereby tightly coupling the external operating conditions with the internal thermal response.

[0042] The specific processing procedure is as follows: First, the reference case temperature value determined in step S1 is retrieved as the starting point for calculation. For the high-temperature trigger threshold, a preset temperature rise offset is added to the reference case temperature value, and the sum is the high-temperature trigger threshold. This temperature rise offset reflects the expected rise in case temperature of critical power devices under typical high-stress conditions. For the low-temperature trigger threshold, a preset temperature drop offset is subtracted from the reference case temperature value, and the difference is the low-temperature trigger threshold. This temperature drop offset ensures that critical power devices obtain sufficient thermal recovery space under low-stress conditions. The two thresholds together constitute the cyclic triggering conditions and are stored for subsequent real-time comparison.

[0043] In one embodiment, during the test preparation phase, the technician reads the reference shell temperature as 85°C from the record, sets the preset temperature rise offset to 20°C, and adds them together to obtain the high temperature trigger threshold of 105°C; sets the preset temperature drop offset to 15°C, and subtracts them to obtain the low temperature trigger threshold of 70°C. Then, these two thresholds are entered into the control software, and the screen displays that the threshold setting is complete.

[0044] S202: Organize start-up, shutdown, and load mutation into a task sequence.

[0045] After the cycle triggering conditions are established, the typical working conditions that are prone to degradation in actual use of marine charging power supplies need to be arranged in a sequence. Each task corresponds to a stress level to ensure that the cycle process fully covers the complete cycle of thermal stress accumulation and recovery, while facilitating the orderly management of the state between nodes.

[0046] The specific processing procedure is as follows: The task sequence is constructed by alternating between high-stress and low-stress tasks. High-stress tasks include starting the charging power supply and applying rated load to the controllable electronic load, or a sudden jump from low load to high load (e.g., from 50% to 100%), to generate significant power conversion heat. Low-stress tasks include shutting down the charging power supply or reducing the load to a low level (e.g., to 10% or no load) or no load, to allow the case temperature of critical power devices to naturally decrease. The task sequence starts with a low-stress task, and is sequentially labeled as Node 1 corresponding to a low-stress task, Node 2 corresponding to a high-stress task, Node 3 corresponding to a low-stress task, and so on, until the cycle ends.

[0047] In one embodiment, the technician defines a sequence in the control program. The first node is set to stop the charging power supply and keep it unloaded. The second node is set to start the power supply and switch the electronic load to full load. The third node returns to stop and unloaded. The program displays that the task sequence has been loaded and is ready to enter automatic execution.

[0048] S203: The task sequence node switching is specified to be driven by the trigger condition of the case temperature of the critical power device crossing.

[0049] The task sequence has been organized. In order to make the cycle rhythm follow the internal thermal response of the equipment completely rather than a fixed time, the node switching rules must be strictly bound to the real-time relationship between the case temperature and threshold of the key power device, so that the length of each cycle can adapt to individual thermal inertia and operating condition differences.

[0050] The specific processing procedure is as follows: The current case temperature of key power components is continuously monitored using temperature sensors. During high-stress tasks, if the current case temperature of a key power component rises to or exceeds the high-temperature trigger threshold, the current node ends and the process switches to the next low-stress task node. Conversely, during low-stress tasks, if the current case temperature of a key power component drops to or below the low-temperature trigger threshold, the current node ends and the process switches to the next high-stress task node. Each switch records the new node sequence number and synchronously triggers subsequent data acquisition steps, forming a complete internal thermal state-driven cycle.

[0051] In one embodiment, after the cycle begins, the equipment is initially in a stopped state, and the shell temperature gradually rises from room temperature. When the temperature sensor shows that the shell temperature reaches 105°C, the control system automatically starts the power supply and switches the load to full load. The panel indicates that it has entered a high-stress node. Subsequently, the shell temperature rises during full-load operation and then cools naturally. When the shell temperature drops to 70°C, the system automatically stops and unloads again. The panel displays that one cycle has been completed and the system has entered the next node.

[0052] Step S2 determines the high-temperature trigger threshold and low-temperature trigger threshold by adding or subtracting a preset offset from the reference case temperature. The start-up and shutdown of the charging power supply and the sudden load change are alternately organized into a task sequence of high-stress tasks and low-stress tasks. It is stipulated that the node switching is triggered when the current value of the case temperature of the key power device is equal to or exceeds the high-temperature trigger threshold, or equal to or lower than the low-temperature trigger threshold. This realizes the cyclic process driven by the internal thermal state, so that the test cycle adapts to the actual thermal response of the equipment and ensures that the node state point corresponds to the natural thermal stress cycle. This provides a consistent internal driving timing basis for the subsequent collection and serialization of health feature values ​​at the node.

[0053] This invention establishes a task sequence node switching mechanism driven by the case temperature of key power devices in step S2. Each node switching moment corresponds to a state point where the internal thermal stress of the device naturally reaches its boundary. These points represent the turning points of typical stress cycles and have high representativeness of degradation. Step S3, based on this, synchronously collects key state information at each node switching point and immediately converts it into a unified offset form relative to the benchmark in step S1 to generate health feature values. This ensures consistent comparability of data across different nodes and different cycle periods. This design avoids the problem of absolute values ​​being affected by environmental or individual differences, allowing health feature values ​​to directly reflect the degree of deviation of the device from the health benchmark. This provides standardized, serialized input data for the subsequent step S4 to extract degradation indices and determine degradation segment boundaries within adjacent node windows, thereby supporting reliable identification of early, continuous degradation.

[0054] The specific implementation method of step S3 is as follows: S301: Synchronously collect the case temperature and output ripple characteristic values ​​of key power devices at each node switch in the task sequence.

[0055] Step S2 has achieved node switching by crossing the high temperature trigger threshold or low temperature trigger threshold of the key power device case temperature. At this moment, the internal thermal stress of the device is at the transition boundary of accumulation or recovery, and the status information is the most representative. Therefore, the acquisition must be strictly bound to the moment of switching to ensure that the data points accurately correspond to the stress cycle transition of the internal drive.

[0056] The specific processing procedure is as follows: Ripple feature value. The current shell temperature and the current output ripple feature value are associated with the current node sequence number and temporarily stored, waiting to be converted into a health feature value later.

[0057] In one embodiment, during cyclic execution, when the temperature sensor displays that the shell temperature has reached the high temperature trigger threshold, the control device automatically switches the load and sends a switching completion signal. The acquisition unit immediately records the shell temperature value of 105°C, and at the same time, the differential probe captures the ripple waveform and calculates the peak value of 140mV. The screen displays that the node data has been captured.

[0058] S302: Relative reference shell temperature and reference ripple generate healthy characteristic values ​​according to a unified offset rule.

[0059] Although the current shell temperature and current output ripple characteristic value directly reflect the node status, the absolute values ​​are easily affected by individual differences in equipment or the temperature of the test environment. In order to achieve a unified comparison of the status between different nodes, the two must be converted into a relative offset ratio relative to the health benchmark in step S1, so as to highlight the deviation caused by the degradation of the equipment itself rather than the interference of external factors.

[0060] The specific processing steps are as follows: When generating the healthy characteristic shell temperature, first subtract the reference shell temperature value from the current shell temperature value to obtain the shell temperature difference value. Then, divide the shell temperature difference value by the reference shell temperature value. The resulting quotient is the healthy characteristic shell temperature. A positive relative offset ratio indicates a temperature rise deviation relative to the healthy reference, while a negative value indicates a temperature drop deviation. The dimensionless characteristic ensures direct comparability under different device reference levels. When generating the healthy characteristic ripple, first subtract the reference ripple value from the current output ripple characteristic value to obtain the ripple difference value. Then, divide the ripple difference value by the reference ripple value. The resulting quotient is the healthy characteristic ripple. A positive relative offset ratio indicates an increased ripple deviation. The healthy characteristic shell temperature and the healthy characteristic ripple are combined into a healthy characteristic value pair for the current node.

[0061] In one embodiment, the current shell temperature is 105°C, the reference shell temperature is 85°C, the difference of 20°C is calculated and divided by 85°C to obtain a healthy characteristic shell temperature of approximately 0.235°C; the current ripple characteristic value is 140mV, the reference ripple is 120mV, the difference of 20mV is calculated and divided by 120mV to obtain a healthy characteristic ripple of approximately 0.167°C, and the software displays that the current node health characteristic value pair has been generated.

[0062] S303: Store the health feature values ​​along with the nodes sequentially as serialized records.

[0063] Once health feature pairs have been generated, in order to support the subsequent extraction of degradation indices between adjacent nodes and the determination of continuous deterioration, all records must be accumulated and saved in node order to form a complete serialized data structure, which facilitates the tracing of the continuous evolution trajectory of the device status.

[0064] The specific processing procedure is as follows: Each node is assigned an incrementally increasing node sequence number, starting from 1 and incrementing sequentially. The current node's node sequence number, health characteristic shell temperature, health characteristic ripple, and optional current shell temperature and current output ripple feature values ​​are packaged into a complete record. This record is then appended to the end of the serialized record list, forming a sequence of data strictly arranged according to node order, which is used in subsequent steps to directly extract adjacent node window data for exponential calculation.

[0065] In one embodiment, after the node switching is completed, the software packages the node sequence number 15, the health characteristic shell temperature 0.235, and the health characteristic ripple 0.167 into a record and appends it to the list. The screen displays that the sequence record has been updated and the total number of nodes has reached 15.

[0066] In step S3, this invention generates a serialized record arranged according to the internal thermally driven nodes, where the health characteristic shell temperature and health characteristic ripple highlight the deviation trend of the device relative to the health benchmark. Step S4 uses this sequence as a basis to extract the shell temperature exceeding the benchmark thermal occupancy index and the ripple spectrum sideband fingerprint index as complementary degradation evidence within adjacent node windows. The dual indices are input into a Gaussian mixture model trained from early healthy samples to obtain a degradation confidence coefficient. Degradation segment boundaries are determined and markers are generated only when the degradation confidence coefficient is consistently high and the health characteristic value continuously shifts simultaneously. This mechanism transforms the original state data into stable, dimensionless degradation trajectory evidence, supporting reliable labeling of continuous degradation stages before device failure, and providing accurate trajectory segmentation for step S5 to estimate the remaining usable history based on the staged change rate.

[0067] In the cycle life test of this invention, what is needed is an input quantity that remains sensitive in the early degradation stage, comparable across different task sequence nodes, and can be stably obtained without increasing the number of acquisition channels. The case temperature of critical power devices directly reflects the thermal stress level experienced by the power devices and their packaging structures during start-up, shutdown, and load abrupt changes. Its over-temperature occupancy relative to the reference can distinguish between short-term spikes and sustained overheating, and align different window durations in a dimensionless form, thus being more suitable as an expression of degradation accumulation trends. In contrast, simple instantaneous temperature, temperature rise, or absolute temperature is more susceptible to transient changes in operating conditions and the external environment, making it difficult to provide stable and consistent degradation evidence under node switching conditions. The sideband energy ratio of the output ripple reflects the structural changes in the power energy conversion and filtering stages. Especially when device parameters drift slowly, the ripple morphology often changes before the final functional indicators. Using a sideband fingerprint rather than the amplitude itself can reduce the interference of load level changes and sampling scale on the results, making it more like an electrical signature related to degradation. The two correspond to thermal stress occupancy and electrical signature changes, respectively. Their information sources are independent and complementary, and they can form more reliable dual evidence inputs for continuous degradation within the same node window. They are suitable for extracting degradation confidence coefficients from the distribution differences between health and degradation using Gaussian mixture models, which can be used for stabilization of subsequent degradation segment boundary determination.

[0068] The specific implementation method of step S4 is as follows: S401: Define adjacent node windows for serialized records and calculate the shell temperature over-reference heat utilization index.

[0069] Step S3 has generated a serialized record arranged in node order. To quantify the continuous cumulative effect of thermal stress in consecutive cycles, time windows must be defined between adjacent nodes, and integration processing should only be performed on the portion where the case temperature of critical power devices exceeds the reference case temperature. This highlights the impact of long-term thermal occupancy on the device structure rather than short-term peak interference. Thermal stress occupancy characteristics include the case temperature exceeding the reference thermal occupancy index.

[0070] The specific processing procedure is as follows: A window between two adjacent nodes is selected from the serialized record. The window starts at the switching time of the previous node and ends at the switching time of the next node, and the duration of the window is recorded. Within the window, the current case temperature of the critical power device is continuously monitored, and only the portion exceeding the reference case temperature is integrated over time: First, at each sampling time, if the current case temperature of the critical power device is higher than the reference case temperature, the difference is taken; otherwise, it is zero. Then, all differences are multiplied by the corresponding sampling time interval and accumulated to obtain the cumulative over-temperature. When calculating the case temperature over-reference heat occupancy index, the cumulative over-temperature is divided by the product of the window duration and the reference case temperature value. The quotient is the case temperature over-reference heat occupancy index, which is a dimensionless value reflecting the average proportion of thermal stress occupancy within the window.

[0071] In one embodiment, during the loop, after the node switch is completed, the control device marks the window from the previous switch time to the current time. During this period, the temperature sensor continuously transmits shell temperature data. The software only calculates the difference for the part that is higher than 85°C and integrates and accumulates it to obtain the over-temperature accumulation. Then, it divides it by the product of the window duration and 85°C to obtain the shell temperature over-reference heat occupancy index of 0.12. The screen displays that the heat occupancy index of this window has been calculated.

[0072] S402: Calculate the ripple spectrum sideband fingerprint index.

[0073] The shell temperature exceeding the reference thermal occupancy index within the same adjacent node window has been obtained. To provide evidence of electrical degradation independent of thermal stress, it is necessary to extract the spectral sideband energy percentage from the output ripple waveform. This allows for the capture of waveform morphology distribution changes caused by filter parameter drift, rather than simple amplitude fluctuations. Ripple spectral morphology characteristics include the ripple spectral sideband fingerprint index.

[0074] The specific processing procedure is as follows: Multiple short-time output ripple waveform segments are extracted within adjacent node windows. Spectral decomposition is performed on each waveform segment, and the sum of spectral energy in a predefined sideband frequency range near the switch's fundamental frequency is extracted as the sideband spectral energy. Simultaneously, the sum of spectral energy across the entire spectrum of that waveform segment is calculated as the total ripple spectral energy. The sideband spectral energy is divided by the total ripple spectral energy to obtain the proportion. Then, the average of the proportions of multiple waveform segments within the window is taken. This average value is the ripple spectral sideband fingerprint index. This dimensionless value highlights the proportion of sideband energy in the total energy as an electrical signature.

[0075] The sideband frequency range is determined symmetrically on both sides of the switch base frequency with a preset bandwidth, or determined according to the combination relationship between the switch base frequency and the modulation frequency.

[0076] In one embodiment, the differential probe within the window captures multiple ripple waveforms. The software performs spectral decomposition on each waveform, extracts the energy of the sideband regions on both sides of the switch's fundamental frequency, divides it by the full spectrum energy to obtain the proportion, and then averages the proportions of multiple segments to obtain the ripple spectrum sideband fingerprint index of 0.18. The screen displays that the electrical fingerprint index of this window has been calculated.

[0077] S403: Input the two exponents into the Gaussian mixture model to obtain the degradation confidence coefficient.

[0078] The shell temperature over-reference heat occupancy index and the ripple spectrum sideband fingerprint index have been extracted. In order to obtain probabilistic evidence from the difference between the healthy and degenerate distributions, it is necessary to train the Gaussian mixture model in advance using early healthy window data and use the double index of the current window as input to calculate the probability of belonging to the degenerate distribution.

[0079] The specific processing procedure is as follows: In the initial stage of testing, the first few adjacent node windows are selected as healthy samples. The shell temperature above-reference heat occupancy index and the ripple spectrum sideband fingerprint index are calculated for each sample. These pairs are used to train a bi-cluster Gaussian mixture model containing healthy and degenerate clusters. After training, the Gaussian component with the lower joint mean of the two indices and a more concentrated distribution is identified as the healthy cluster, and the other component is identified as the degenerate cluster. The degradation confidence coefficient is the posterior probability that the current input pair belongs to the degenerate cluster. For the current adjacent node window, the shell temperature above-reference heat occupancy index and the ripple spectrum sideband fingerprint index are combined into an input pair, substituted into the trained model to calculate the posterior probability of falling into the degenerate cluster. The obtained probability value is the degradation confidence coefficient, which is between 0 and 1, directly representing the confidence that the current window belongs to the degenerate distribution.

[0080] In one embodiment, the construction, training, and application of the Gaussian mixture model are described below to ensure that the model can effectively learn the difference between the healthy distribution and the potential degradation distribution from early healthy samples, and to provide stable degradation confidence coefficient calculations for the bi-exponential input of subsequent windows.

[0081] The Gaussian mixture model employs a mixture structure of two Gaussian components: one corresponding to the healthy cluster and the other to the degenerate cluster. Each component is a two-dimensional Gaussian distribution, corresponding to the input feature pair of the shell temperature super-benchmark heat occupancy index and the ripple spectral sideband fingerprint index. The model covariance matrix is ​​set to full covariance form, meaning each component has an independent full covariance matrix, allowing for flexible capture of the correlation between the two features.

[0082] The training data comes from the first 20 adjacent node windows in the initial testing phase. These windows are generated when the device is brand new or in good condition. The calculated shell temperature over-reference heat occupancy index and ripple spectrum sideband fingerprint index are used as sample points, totaling 20 sample pairs. The expectation-maximization algorithm is used for parameter estimation during the training process. Initialization is performed first: the K-means clustering algorithm is used to initially cluster the 20 sample pairs, dividing the samples into two clusters. One cluster is labeled as the healthy cluster (usually the cluster with lower index values), and the other cluster is labeled as the degenerate cluster (the cluster with higher index values). This yields the initial mean vector, covariance matrix, and mixing coefficients (the initial mixing coefficients are all set to 0.5).

[0083] The process then proceeds to the expectation-maximization iteration: In the forward step, for each sample pair, the posterior probability of it belonging to the healthy cluster and the degenerate cluster is calculated; in the maximization step, the mean vector, covariance matrix, and mixing coefficients of each cluster are updated based on the posterior probabilities. The iteration termination condition is set to the log-likelihood value changing by less than 0.001 between two consecutive iterations, or reaching the maximum number of iterations (300), to ensure convergence stability. After training, the model parameters, including the mean vector, covariance matrix, and mixing coefficients of the healthy and degenerate clusters, are fixed.

[0084] For each new adjacent node window during the testing process, the calculated shell temperature over-reference heat occupancy index and ripple spectrum sideband fingerprint index are used as input feature vectors and substituted into the trained Gaussian mixture model. First, the probability density of the input vector in the healthy and degenerate clusters is calculated. Then, the total probability density is calculated by combining the mixing coefficients. Finally, the posterior probability that the input vector belongs to the degenerate cluster, i.e., the degradation confidence coefficient, is calculated. Specifically, the posterior probability is derived using Bayes' theorem: the degradation cluster probability density multiplied by the degradation cluster mixing coefficient, divided by the total probability density (the weighted sum of the probability densities of the healthy and degenerate clusters). The resulting degradation confidence coefficient ranges from 0 to 1 and is directly used for subsequent degradation segment boundary determination.

[0085] In this embodiment, the technicians integrated an open-source machine learning library (such as the GaussianMixture class in scikit-learn) into the test software, setting n_components=2, covariance_type='full', max_iter=300, tol=0.001, and init_params='kmeans'. The first 20 windows of bi-exponential sample pairs were loaded and the fit method was executed to complete the training. Subsequently, the predict_proba method was executed on the new window of bi-exponential samples to extract the probability of belonging to the degenerate cluster as the degeneracy confidence coefficient. The software interface displayed that the model training was complete and output the degeneracy confidence coefficient value of each window in real time.

[0086] S404: Determine the boundary of the degradation segment and generate a degradation segment identifier when the degradation confidence coefficient is consistently high and the health feature value is continuously shifted.

[0087] The degradation confidence coefficient and exponent of a single window may be affected by occasional disturbances. To ensure that the identification device enters the continuous degradation stage, it is necessary to monitor the consistent performance of multiple consecutive windows in the sequence and set synchronization constraints in combination with the offset direction of health feature values.

[0088] The specific processing procedure is as follows: traverse adjacent node windows in the serialized record, check whether the degradation confidence coefficients of multiple consecutive windows all exceed the preset high-order threshold, and simultaneously check whether the health feature shell temperature and health feature ripple within the same consecutive window both show a unidirectional continuous shift, for example, both continuously increasing in a positive direction. When the high-order continuous degradation confidence coefficient and the unidirectional shift of the health feature value are simultaneously satisfied, it is determined that the degradation segment boundary is established from the node corresponding to the first satisfied window. Subsequently, a degradation segment identifier is generated, marking that the node and subsequent nodes belong to the degradation segment, until the condition is no longer satisfied or the test ends.

[0089] In one embodiment, after the loop progresses to a certain number of nodes, the software checks that the degradation confidence coefficients of the five most recent windows all exceed 0.75, and that the shell temperature and ripple of the health feature both continue to increase. It determines that the degradation segment boundary is established, marks the degradation segment from the corresponding node, and displays on the screen that the degradation segment has been marked. The test continues to execute until the termination condition is met.

[0090] Step S4 obtains the over-reference thermal occupancy index by normalizing the portion of the critical power device case temperature higher than the reference case temperature within the adjacent node window of the serialized record after time integration. The ripple spectrum sideband fingerprint index is obtained by the energy ratio of the short-time waveform spectrum sideband of the ripple. The dual indices are input into the Gaussian mixture model trained on the early healthy samples to calculate the degradation confidence coefficient. When the degradation confidence coefficient and the healthy characteristic case temperature and healthy characteristic ripple are continuously synchronously high, the degradation segment boundary is determined to be established and a degradation segment identifier is generated. This transforms the serialized record into a degradation trajectory with stage annotations, ensuring that the subsequent step S5 can calculate the remaining usable history and export the complete degradation trajectory data package based on the continuous history and change rate of the reliable degradation segment when the device is functioning normally.

[0091] This invention constructs a complete degradation trajectory through steps S1 to S4: from establishing a health baseline, internal thermal-driven cycling, and serializing health feature values ​​at nodes, to dual-exponential extraction and Gaussian mixture model determination of adjacent node windows, ultimately generating a reliable degradation segment identifier. This identifier clearly divides the trajectory into healthy and degraded segments, highlighting the starting point and evolution process of continuous degradation. Step S5, based on this staged trajectory, quantifies the rate of change of health feature values ​​within the degradation segment and extrapolates this rate to the remaining usable time to a preset failure threshold, thereby transforming the lifespan conclusion from an end-point judgment of whether failure has occurred to a quantifiable prediction of how long it can still operate stably. Simultaneously, a complete data package containing all original and derived parameters is exported, directly available for external fault prediction and health management systems, realizing the transformation of marine charging power supplies from passive emergency repairs to planned maintenance.

[0092] The specific implementation method of step S5 is as follows: S501: Determine the duration of the degradation segment based on the degradation segment identifier and calculate the rate of change of health characteristic values.

[0093] Step S4 has divided the serialized record into healthy and degraded segments by degraded segment identifiers. In order to accurately quantify the evolution rate of the device after it enters continuous degradation, the continuous history of the degraded segment must be calculated first, and the cumulative offset of the healthy characteristic value is normalized by this history, so as to obtain a dimensionless speed index that reflects the speed of degradation.

[0094] The specific processing steps are as follows: Extract the node subsequence corresponding to the degradation segment identifier from the serialized record, and count the number of nodes contained in this subsequence as the degradation segment's duration. Calculate the cumulative offset of the healthy feature shell temperature within the degradation segment. First, subtract the healthy feature shell temperature value of the starting node of the degradation segment from the healthy feature shell temperature value of the terminal node of the degradation segment to obtain the total shell temperature offset. Then, divide the total shell temperature offset by the degradation segment's duration to obtain the healthy feature shell temperature change rate, which represents the average shell temperature offset increment for each node. Similarly, calculate the healthy feature ripple change rate. First, subtract the healthy feature ripple value of the starting node of the degradation segment from the healthy feature ripple value of the terminal node of the degradation segment to obtain the total ripple offset. Then, divide the total ripple offset by the degradation segment's duration to obtain the healthy feature ripple change rate. Take the larger value between the healthy feature shell temperature change rate and the healthy feature ripple change rate as the comprehensive change rate, ensuring that the overall degradation rate is conservatively represented using the fastest degradation dimension.

[0095] In one embodiment, when the test ends, the software extracts the subsequence marked as the degenerate segment starting from node 150 from the serialized record, counts the number of nodes as 80, calculates the final health characteristic shell temperature of 0.75 minus the initial 0.35 to obtain the total offset of 0.40, divides it by 80 to obtain the shell temperature change rate of 0.005, calculates the ripple change rate as 0.003, takes the larger value of 0.005 as the comprehensive change rate, and displays on the screen that the degenerate segment change rate has been calculated.

[0096] S502: Calculate the remaining available time based on the comprehensive rate of change.

[0097] With the overall rate of change obtained, in order to predict the additional capacity required for the equipment to go from its current degradation state to functional failure, a clear failure threshold must be set, and a linear extrapolation assuming a consistent degradation rate must be performed to obtain the remaining available history as a quantitative window for planned maintenance.

[0098] The specific processing procedure is as follows: A preset failure threshold is used as the boundary when the relative offset of the health characteristic value reaches the limit of performance margin exhaustion. For example, a relative offset of 1.0 corresponds to a key indicator exceeding the design tolerance. The larger value between the shell temperature and ripple of the health characteristic at the end node of the degradation segment is used as the starting point of the current health characteristic value. The remaining offset capacity is obtained by subtracting the starting point of the current health characteristic value from the failure threshold. Then, the remaining offset capacity is divided by the overall rate of change, and the quotient is the remaining available history. This history represents the number of additional nodes required from the current node to reach the failure threshold. If the remaining offset capacity is negative or the remaining available history is less than the preset minimum safety margin, the remaining available history is directly set to zero to avoid overly optimistic predictions.

[0099] In one embodiment, the software reads a preset failure threshold of 1.0, takes the current health characteristic value as 0.75, subtracts 0.75 from 1.0 to get the remaining offset capacity of 0.25, divides it by the comprehensive change rate of 0.005 to get the remaining available process of 50 nodes, and displays on the screen that the remaining available process has been calculated, which is equivalent to an additional 50 cycle periods.

[0100] S503: Output cycle life test results and export degradation trajectory data package.

[0101] The remaining available time has been estimated. In order to complete the entire cycle life test and support the predictive maintenance of marine charging power supplies, it is necessary to integrate the total time, degradation segment information and prediction results to form a conclusion. At the same time, all raw and derived data should be packaged to form a complete trajectory, which can be directly called by the external fault prediction and health management system.

[0102] The specific processing procedure is as follows: The cycle life test conclusion includes the total number of cycle nodes, the starting node position of the degradation segment, the comprehensive rate of change value, the remaining usable history, and the predicted remaining stable operating time, if any time conversion coefficient is used. The exported degradation trajectory data package is a structured file, covering the case temperature, output ripple characteristic value, health characteristic case temperature, health characteristic ripple, case temperature exceeding the reference thermal occupancy index, ripple spectrum sideband fingerprint index, degradation confidence coefficient, and degradation segment identification mark for all nodes in sequence. The data package includes test conclusion text and failure threshold descriptions to ensure that the file can be directly used for subsequent model training or state monitoring after loading.

[0103] In one embodiment, at the end of the test, the software generates a conclusion report showing a total of 230 nodes, a degradation segment starting from node 150, an overall change rate of 0.005, and 50 remaining available nodes. Then, all node data and identifiers are packaged into a CSV file. Technicians copy the file to the health management system, and the screen displays that the test conclusion has been output and the data package has been exported.

[0104] Step S5 calculates the comprehensive change rate by dividing the cumulative offset of the health characteristic shell temperature and health characteristic ripple based on the degradation segment identifier by the duration of the degradation segment. The remaining usable time is calculated by subtracting the current health characteristic value from the preset failure threshold and dividing by the comprehensive change rate. The cycle life test conclusion, which includes the total time, degradation segment location, change rate, and remaining usable time, is output. A degradation trajectory data package covering the shell temperature of key power devices, output ripple characteristic value, health characteristic shell temperature, health characteristic ripple, shell temperature over-reference thermal occupancy index, ripple spectrum side-band fingerprint index, degradation confidence coefficient, and degradation segment identifier is exported. This allows the test conclusion to directly support fault prediction and health management applications, enabling advance planning and engineering implementation of marine charging power supply maintenance decisions.

[0105] Specifically, the above are merely preferred embodiments of this application and are not intended to limit this application.

[0106] The aforementioned thresholds, such as the shell temperature stability threshold, preset high threshold, or preset failure threshold, can be pre-calibrated through offline simulation testing or set to fixed values ​​according to on-site operating procedures.

[0107] In the description of this specification, references to terms such as "an embodiment," "an example," and "a specific example" indicate that a particular feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0108] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for testing the cycle life of a marine charging power supply, characterized in that, Including the following steps: S1: Connect the marine charging power supply under test to the standard input and controllable load for pre-running. After the case temperature of the key power device stabilizes, collect the case temperature and output ripple characteristic values ​​and set them as the reference case temperature and reference ripple, respectively. S2: Set the cycle triggering conditions with the reference case temperature, construct the task sequence of start-stop and load change, and drive node switching across the triggering conditions with the case temperature of key power devices to form a cycle driven by internal thermal state. S3: Synchronously collect the case temperature and output ripple characteristic values ​​of key power devices at each node switching point, generate health characteristic values ​​relative to the reference case temperature and reference ripple according to a unified offset rule, and store them as serialized records in sequence with the node order. S4: Extract thermal stress occupancy features and ripple spectrum morphology features from adjacent node windows, input the two types of features into a pre-trained model trained by healthy segment samples to obtain degradation confidence coefficients, and determine the degradation segment boundary and generate degradation segment identifiers when the degradation confidence coefficients continuously exceed the standard and the healthy feature values ​​continue to shift. S5: Based on the continuous history corresponding to the degradation segment identifier and the rate of change of health characteristic values, calculate the remaining usable history and output the lifetime conclusion. At the same time, export the degradation trajectory data package containing shell temperature, ripple, health characteristic values ​​and degradation confidence coefficient for health management.

2. The method for testing the cycle life of a marine charging power supply according to claim 1, characterized in that, Step S1 includes: Connect the marine charging power supply under test to the standard input power supply and the controllable electronic load, and start the pre-run with a constant load; monitor the case temperature of key power devices in real time, calculate the absolute difference of case temperature between adjacent sampling times, and determine that it has entered the stable range after it is continuously less than the preset condition; select a continuous acquisition window from the starting point of the stable range, and continuously record the case temperature of key power devices and the peak-to-peak value of output ripple; weight and accumulate the case temperature of key power devices in the window according to the sampling time interval and divide it by the duration of the window to obtain the reference case temperature, and determine the maximum value of the peak-to-peak value of output ripple in the window as the reference ripple.

3. The method for testing the cycle life of a marine charging power supply according to claim 2, characterized in that, Step S2 includes: The high-temperature trigger threshold is determined by adding a preset temperature rise offset to the reference case temperature, and the low-temperature trigger threshold is determined by subtracting a preset temperature drop offset from the reference case temperature. The charging power supply start-up and shutdown and load change are alternately organized into a task sequence of high-stress tasks and low-stress tasks.

4. The method for testing the cycle life of a marine charging power supply according to claim 3, characterized in that, Step S2 also includes: The task sequence node switching is triggered when the current case temperature of the critical power device is equal to or exceeds the high temperature trigger threshold, switching from a high stress task to a low stress task; and when the current case temperature of the critical power device is equal to or below the low temperature trigger threshold, switching from a low stress task to a high stress task.

5. A method for testing the cycle life of a marine charging power supply according to claim 4, characterized in that, Step S3 includes: At the moment of triggering node switching and before the switching action is executed, the case temperature and output ripple characteristic values ​​of key power devices are collected synchronously. The healthy characteristic case temperature is generated by subtracting the reference case temperature from the current case temperature and dividing by the reference case temperature. The healthy characteristic ripple is generated by subtracting the reference ripple from the current output ripple characteristic value and dividing by the reference ripple. The data are stored sequentially as serialized records along with the node.

6. A method for testing the cycle life of a marine charging power supply according to claim 5, characterized in that, Step S4 includes: The over-temperature accumulation is obtained by integrating the portion of the critical power device case temperature that is higher than the reference case temperature within the adjacent node window of the serialized record over time. Then, the over-temperature accumulation is divided by the product of the window duration and the reference case temperature to obtain the case temperature over-reference heat occupancy index.

7. A method for testing the cycle life of a marine charging power supply according to claim 6, characterized in that, Step S4 also includes: The short-time output ripple waveform within the adjacent node window is decomposed into a spectrum to extract the sideband spectrum energy near the switching fundamental frequency. The sideband fingerprint index of the ripple spectrum is obtained by dividing it by the total ripple spectrum energy. The shell temperature over-reference heat occupancy index and the sideband fingerprint index of the ripple spectrum are input into the Gaussian mixture model trained on the healthy segment samples to obtain the degradation confidence coefficient.

8. A method for testing the cycle life of a marine charging power supply according to claim 7, characterized in that, Step S4 also includes: When the degradation confidence coefficient of multiple consecutive adjacent node windows continuously exceeds the preset high-order threshold and the health feature value continues to shift, the degradation segment boundary is determined to be established and a degradation segment identifier is generated.

9. A method for testing the cycle life of a marine charging power supply according to claim 8, characterized in that, Step S5 includes: Based on the degraded segment identifier, extract the node subsequence of the degraded segment, calculate the difference in shell temperature and ripple of the healthy feature between the end and the beginning of the degraded segment, and divide them by the number of nodes corresponding to the degraded segment to obtain the rate of change. Take the larger of the two as the comprehensive rate of change.

10. A method for testing the cycle life of a marine charging power supply according to claim 9, characterized in that, Step S5 also includes: The remaining usable history is obtained by subtracting the maximum value of the health characteristics at the end of the degradation segment from the preset failure threshold and dividing by the comprehensive change rate. The test conclusion is output, which includes the total number of nodes, the location of the degradation segment, the comprehensive change rate and the remaining usable history. The degradation trajectory data package containing the corresponding parameters of all nodes and the degradation segment identifier is exported.

Images