A method for identifying degradation trends in thermal cycling simulation results of embedded power modules

Abstract

This invention provides a method for identifying the degradation trend of thermal cycling simulation results of embedded power modules. The method includes: reading the control allocation record, sensor sampling record, and thermal cycling simulation results of the target embedded power module to generate an actual power allocation sequence, an actual boundary condition sequence, and an actual thermal cycling feature sequence; selecting healthy control allocation samples based on the actual boundary condition sequence, generating a reference power allocation sequence using imitation learning, and generating a load-yielding compensation sequence by combining the actual power allocation sequence; constructing an equal-load reference thermal cycling simulation input set based on the reference power allocation sequence and the actual boundary condition sequence under the condition that the fixed structural parameter set remains unchanged, optimizing the reference thermal cycling simulation parameters using the bat algorithm, and generating an equal-load reference thermal cycling feature sequence; and generating a compensation difference sequence based on the actual thermal cycling feature sequence and the equal-load reference thermal cycling feature sequence.

Description

Technical Field

[0001] This invention relates to the field of power module thermal state analysis and degradation identification technology, and in particular to a method for identifying degradation trends from thermal cycling simulation results of embedded power modules. Background Technology

[0002] Embedded power modules are widely used in new energy vehicle drive systems, charging devices, and industrial power electronic equipment. During long-term operation, these modules repeatedly experience load fluctuations and changes in heat dissipation conditions, resulting in continuous thermal cycling. Thermal cycling causes thermal stress accumulation in the chip layer, package layer, and interface layer, leading to solder fatigue, interface aging, and a decrease in localized heat dissipation capacity. Therefore, identifying degradation trends in embedded power modules based on thermal cycling simulation results has become a common technique in design verification, lifespan assessment, and fault prediction.

[0003] Existing technologies typically rely on temperature cycling amplitude, thermal resistance change, and temperature rise slope obtained from thermal cycling simulations to determine the degradation state of power modules. This method is applicable to scenarios where a single module operates independently and the load distribution is relatively fixed. However, in practical applications, embedded power modules often operate in multi-phase parallel, current-sharing, or power redistribution environments, where the controller dynamically adjusts the power load on each module. When an embedded power module begins to degrade, the control system usually reduces the power it bears and transfers some of the load to other modules, causing the module's actual thermal response to stabilize. In this situation, relying solely on actual thermal cycling simulation results can easily misjudge a module that has already undergone internal degradation but is protected by load reduction as having improved condition or experiencing slower degradation.

[0004] Therefore, this invention proposes a method for identifying degradation trends from thermal cycling simulation results of embedded power modules. The information disclosed in the background section is only for enhancing understanding of the background of this disclosure and may therefore contain prior art information that is not common knowledge to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for identifying degradation trends in thermal cycling simulation results of embedded power modules, thereby solving the technical problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for identifying degradation trends in thermal cycling simulation results of embedded power modules includes the following steps: S1. Obtain the control allocation record, sensor sampling record and thermal cycling simulation result of the target embedded power module, perform time alignment, period slicing and module mapping on the record, and generate the actual power allocation sequence, actual boundary condition sequence and actual thermal cycling characteristic sequence including temperature cycle amplitude, period thermal resistance change and period temperature rise slope. S2. Based on the actual boundary condition sequence, select health control allocation samples from the healthy operation sample library, use imitation learning to generate the reference power allocation sequence of the target embedded power module in the healthy state, and combine it with the actual power allocation sequence to generate the load-yielding compensation sequence. S3. Under the condition that the fixed structural parameter set remains unchanged, construct the equal load reference thermal cycle simulation input set based on the reference power distribution sequence and the actual boundary condition sequence, optimize the reference thermal cycle simulation parameters using the bat algorithm, execute the reference thermal cycle simulation, and generate the equal load reference thermal cycle feature sequence. S4. Construct a comparison sample set based on the actual thermal cycle characteristic sequence, the equal load reference thermal cycle characteristic sequence, and the load-yielding compensation sequence, and generate a compensation difference sequence and a pseudo-health judgment index sequence. S5. Based on the pseudo-health judgment index sequence, actual thermal cycle characteristic sequence, equal load reference thermal cycle characteristic sequence and load-bearing compensation sequence, extract pseudo-health trend segments, generate staged degradation judgment results, and output the degradation trend identification results of the target embedded power module.

[0007] S1 specifically includes: reading the controller log stream, smart sensor sampling stream, and thermal cycling simulation result file corresponding to the target embedded power module to form the original running dataset; performing time alignment, cycle slicing, and module mapping on the original running dataset, deleting data segments with missing timestamps, abnormal self-test status, mismatched operating condition identifiers, and non-closed cycles, and generating a set of periodic running data blocks; extracting actual power allocation information, actual boundary condition information, and periodic thermal response information based on the set of periodic running data blocks, and generating actual power allocation sequence, actual boundary condition sequence, and actual thermal cycling feature sequence.

[0008] S2 specifically includes: reading the healthy operation sample library and the actual boundary condition sequence, verifying the integrity of the sample fields, and filtering healthy samples according to the operating condition distance between the cooling medium inlet temperature, cooling medium flow rate, fan speed, and the duration of the control allocation segment and the target cycle, generating a healthy control allocation sample set; based on the healthy control allocation sample set and the actual boundary condition sequence, establishing a benchmark power allocation generation model using imitation learning, and outputting a benchmark power allocation sequence; matching the benchmark power allocation sequence with the actual power allocation sequence according to the cycle number, calculating the load-bearing compensation value for each cycle, and generating a load-bearing compensation sequence by sorting according to the cycle number.

[0009] S3 specifically includes: reading the reference power allocation sequence, the actual boundary condition sequence, and the fixed structural parameter set; constructing an equal-load reference thermal cycle simulation input set while keeping the fixed structural parameter set unchanged; using the bat algorithm to search and optimize the reference thermal cycle simulation parameters based on the equal-load reference thermal cycle simulation input set, and outputting the optimized reference thermal cycle simulation parameter set; performing reference thermal cycle simulation according to the optimized reference thermal cycle simulation parameter set, and generating an equal-load reference thermal cycle feature sequence according to the same feature extraction rules as the actual thermal cycle feature sequence.

[0010] S4 specifically includes: reading the actual thermal cycle characteristic sequence, the equal-load reference thermal cycle characteristic sequence, and the load-yielding compensation sequence; verifying the consistency of module identifier, cycle number, and validity identifier; deleting any missing corresponding cycle record for any object; constructing a comparison sample set according to the same cycle number; calculating the temperature cycle amplitude difference, cycle thermal resistance change difference, and cycle temperature rise slope difference between the reference value and the actual value based on the comparison sample set, and generating a compensation difference sequence after normalization; and performing a fusion calculation based on the compensation difference sequence and the load-yielding compensation value to generate a pseudo-health judgment index sequence sorted by cycle.

[0011] S5 specifically includes: reading the pseudo-health judgment index sequence, the actual thermal cycle characteristic sequence, the equal load reference thermal cycle characteristic sequence, and the load-yielding compensation sequence; extracting pseudo-health trend segments using a sliding window; and calculating the number of duration periods, the growth rate of the index, and the degree of load-yielding compensation enhancement for each pseudo-health trend segment; classifying and judging based on the number of duration periods, the growth rate of the index, and the degree of load-yielding compensation enhancement for each pseudo-health trend segment, and generating a staged degradation judgment result; and generating a degradation trend identification result containing module identifier, load-yielding pseudo-health status identifier, corresponding period interval, stage level, and risk warning information based on the staged degradation judgment result.

[0012] The beneficial effects of this invention are as follows: This invention introduces the actual thermal cycle characteristic sequence, the reference power distribution sequence, the equal load reference thermal cycle characteristic sequence, and the load shedding compensation sequence simultaneously. It no longer judges the degradation state solely based on the actual thermal cycle simulation results. Instead, it can identify pseudo-health phenomena caused by control compensation and power redistribution, thereby solving the problem that existing technologies have difficulty distinguishing between the true health state and the state masked by load shedding.

[0013] This invention utilizes a health control allocation sample set and imitation learning to generate a baseline power allocation sequence, enabling the reconstruction of the power tasks that the target embedded power module should undertake in a healthy state. This provides a unified reference for subsequent load-bearing compensation value calculation and degradation difference identification, thereby improving the comparability and stability of degradation trend identification. Under the condition of keeping the fixed structural parameter set unchanged, an equal-load reference thermal cycle simulation input set is constructed, and the reference thermal cycle simulation parameters are optimized using the bat algorithm to ensure that the reference thermal cycle response is consistent with the actual boundary conditions. This allows for the separation of control compensation factors from structural degradation factors, improving the accuracy of thermal response difference analysis.

[0014] This invention constructs a compensation difference sequence and a pseudo-health assessment index sequence, jointly analyzing differences in temperature cycle amplitude, changes in periodic thermal resistance, differences in periodic temperature rise slope, and load-bearing compensation behavior. This extends the identification of single-cycle differences to continuous-cycle trend identification, improving the ability to detect early-stage, hidden degradation. Furthermore, pseudo-health trend segments are extracted to generate staged degradation assessment results and degradation trend identification results, outputting corresponding cycle intervals, stage levels, and risk warning information. This provides a directly applicable basis for maintenance decisions, control strategy corrections, and screening / removal processes for embedded power modules. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a method for identifying the degradation trend of embedded power module thermal cycling simulation results according to the present invention. Detailed Implementation

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

[0017] Example: Figure 1 As shown in the figure, this embodiment provides a method for identifying the degradation trend of embedded power module thermal cycling simulation results, including the following steps: S1. Obtain the control allocation record, sensor sampling record and thermal cycling simulation result of the target embedded power module, perform time alignment, period slicing and module mapping on the record, and generate the actual power allocation sequence, actual boundary condition sequence and actual thermal cycling characteristic sequence including temperature cycle amplitude, period thermal resistance change and period temperature rise slope. S2. Based on the actual boundary condition sequence, select health control allocation samples from the healthy operation sample library, use imitation learning to generate the reference power allocation sequence of the target embedded power module in the healthy state, and combine it with the actual power allocation sequence to generate the load-yielding compensation sequence. S3. Under the condition that the fixed structural parameter set remains unchanged, construct the equal load reference thermal cycle simulation input set based on the reference power distribution sequence and the actual boundary condition sequence, optimize the reference thermal cycle simulation parameters using the bat algorithm, execute the reference thermal cycle simulation, and generate the equal load reference thermal cycle feature sequence. S4. Construct a comparison sample set based on the actual thermal cycle characteristic sequence, the equal load reference thermal cycle characteristic sequence, and the load-yielding compensation sequence, and generate a compensation difference sequence and a pseudo-health judgment index sequence. S5. Based on the pseudo-health judgment index sequence, actual thermal cycle characteristic sequence, equal load reference thermal cycle characteristic sequence and load-bearing compensation sequence, extract pseudo-health trend segments, generate staged degradation judgment results, and output the degradation trend identification results of the target embedded power module.

[0018] S1 specifically includes the following sub-steps: S110. First, read the acquisition configuration table of the operating system where the target embedded power module is located, and verify whether the acquisition configuration table contains the module identifier, control channel identifier, smart sensor identifier, and simulation condition identifier at the same time. If any field is missing, delete the acquisition configuration and stop the data access to the corresponding module. The reason is that the subsequent module mapping and condition correspondence cannot be completed when the above fields are missing.

[0019] After verification, the controller log stream, smart sensor sampling stream, and thermal cycling simulation result file are connected to form the original running dataset. The original running dataset is divided into three basic records: control allocation records, sensor sampling records, and simulation result records.

[0020] The control allocation record is used to characterize the task allocation status of the controller to the target embedded power module, and includes at least the record timestamp, system identifier, module identifier, control channel identifier, controller-issued power allocation value, controller-issued current allocation value, switching frequency setting value, and protection status identifier.

[0021] The sensor sampling records are used to characterize the measured operating status of the target embedded power module, including at least the sampling timestamp, module identifier, current value, voltage value, case temperature value, cooling medium inlet temperature, cooling medium flow rate, fan speed, and self-test status value.

[0022] The simulation results record is used to characterize the thermal cycle simulation output corresponding to the module, and includes at least the simulation condition identifier, module identifier, cycle number, temperature time series point set, cycle maximum temperature, and cycle minimum temperature. To ensure subsequent time alignment and execution, this embodiment uniformly uses the timestamp recorded in the control allocation record as the primary time reference, and requires that both the sensor sampling record and the simulation results record retain millisecond-level time stamps.

[0023] For example, when the controller log period is 10ms and the smart sensor sampling period is 5ms, they are still uniformly linked to the 10ms main time base. The output of S110 is the raw running dataset, which is read in S120 and time-aligned, period-sliced, and module-mapped.

[0024] S120. First, read the original running dataset and verify whether the association keys of the three types of basic records are complete. In this embodiment, the module identifier, simulation condition identifier and timestamp interval are defined as joint association keys. Only when the joint association keys are established at the same time can the control allocation record, sensor sampling record and simulation result record be allowed to enter the same cycle running data block.

[0025] For records with missing timestamps, missing module identifiers, missing simulation condition identifiers, or abnormal self-test status values ​​of smart sensors, they are directly deleted and do not enter the alignment process; here, abnormal self-test status value means that the self-test status value is not equal to the preset valid value 1.

[0026] After the initial deletion, the remaining records are time-aligned. During time alignment, the timestamp of the control allocation record is used as the alignment axis, and the sensor sampling records and simulation result records are resampled to the same time scale. If there are no valid sensor sampling records for three consecutive main time scales within the same control allocation cycle, that cycle is deleted because the thermal response of that cycle cannot be reliably reconstructed. After time alignment, cycle slicing is performed.

[0027] In this embodiment, a cycle running data block is defined as a closed data segment that starts from the beginning of the current control allocation segment of the target embedded power module and ends when the control allocation segment ends and the case temperature drops to the end of the stable interval before the start of the next control allocation segment.

[0028] If a data segment only shows a heating process but not a corresponding cooling process, then the data segment is deleted because it cannot be used for subsequent extraction of temperature cycle amplitude and periodic temperature rise slope.

[0029] Then, module mapping is performed to verify whether a module identifier uniquely corresponds to a control channel identifier within the same time interval. If a module identifier corresponds to multiple control channel identifiers, all records within that time interval are deleted to prevent the load of other modules from being mistakenly written into the target embedded power module.

[0030] After the above processing is completed, a set of periodic running data blocks is output. Each periodic running data block contains at least one control allocation record, one closed sensor sampling segment, and one matching simulation result segment. This set is directly read in S130 and used to generate all subsequent basic sequences.

[0031] S130. First, read the set of periodic operation data blocks and verify whether each periodic operation data block contains current value, voltage value, shell temperature value, cooling medium inlet temperature, cooling medium flow rate and fan speed at the same time. If any of the above key fields are missing, delete the corresponding periodic operation data block. The reason is that the period cannot support the generation of actual power allocation sequence and actual boundary condition sequence at the same time.

[0032] For the retained periodic running data blocks, the actual power allocation sequence is first generated. Specifically, for the... Power calculations are performed on all valid sampling points within the current period to obtain the power of the first period. Actual power distribution value for each cycle: ; in, For the first The actual power distribution value for each cycle, For the first Number of valid sampling points within a period For the first The first cycle Voltage values ​​at each valid sampling point For the first The first cycle The current value of each valid sampling point.

[0033] The reason for using this average power definition is that the target embedded power module has switching fluctuations within a thermal cycle. Directly taking the value of a single point will amplify the instantaneous disturbance, while the periodic power generated by the average of the effective sampling points is more suitable for subsequent comparison with the baseline power allocation sequence.

[0034] For example, when the number of effective sampling points in a certain period is 20, the voltage value is stable at around 600V, and the current value fluctuates between 48A and 52A, the obtained This represents the actual power allocation for that cycle. After sorting by cycle number, the actual power allocation sequence is output.

[0035] The actual boundary condition sequence is then generated. Specifically, the cooling medium inlet temperature, cooling medium flow rate, fan speed, and control allocation segment duration are read from the same cycle's running data block and written into the boundary condition record according to the cycle number, forming the actual boundary condition sequence. The control allocation segment duration is written in here because the subsequent reference power allocation generation model needs to read not only the heat dissipation boundary but also the duration of that boundary.

[0036] For example, if the inlet temperature of the cooling medium is 28℃, the flow rate of the cooling medium is 12L / min, the fan speed is 2800r / min, and the duration of the control distribution section is 1.6s, then the above four items together constitute a boundary condition record for that cycle.

[0037] Finally, the actual thermal cycle feature sequence is generated. In this embodiment, feature extraction is performed on the shell temperature time series in each cycle's running data block, generating at least the temperature cycle amplitude, the cycle thermal resistance change, and the cycle temperature rise slope. The temperature cycle amplitude is calculated using the following formula: ; in, For the first Temperature cycle amplitude per cycle, For the first The highest temperature of each cycle, For the first The lowest temperature of each cycle. The cycle temperature rise slope is calculated using the following formula: ; in, For the first The periodic temperature rise slope of each cycle, For the first The starting temperature of each heating cycle segment For the first The start time of each cycle of temperature increase For the first The moment when the highest temperature of the cycle is reached. The change in thermal resistance during the cycle is obtained by the difference between the equivalent thermal resistance of the current cycle and the equivalent thermal resistance of the previous cycle, and is written in the order of adjacent cycles.

[0038] After feature extraction is completed, the actual thermal cycle feature sequence is output according to the cycle number. At this point, S130 simultaneously outputs the actual power allocation sequence, the actual boundary condition sequence, and the actual thermal cycle feature sequence. Among them, the actual power allocation sequence is used in S230 to generate the load-bearing compensation sequence; the actual boundary condition sequence is used in S210 to screen health control allocation samples; in S220, it is used to generate the reference power allocation sequence; and in S310, it is used to construct the equal-load reference thermal cycle simulation input set. The actual thermal cycle feature sequence is used in S410-S430 and S510-S530 to generate the compensation difference sequence, the pseudo-health judgment index sequence, and the degradation trend identification result.

[0039] S2 specifically includes the following sub-steps: S210. First, read the healthy operation sample library and the actual boundary condition sequence, and verify whether the sample records in the healthy operation sample library contain at least the module identifier, cycle number, cooling medium inlet temperature, cooling medium flow rate, fan speed, control allocation segment duration, power allocation value, protection status identifier, and degradation status identifier. If any of the above fields are missing, delete the corresponding sample record, because the sample cannot be used for both operating condition matching and demonstration behavior extraction at the same time.

[0040] The "healthy" status in the healthy operation sample library is defined according to fixed admission rules: there are no maintenance or replacement records, no abnormal shutdown records, no continuous over-limit records of shell temperature, no continuous over-limit records of thermal resistance drift, and the protection status indicator shows that the load shedding protection has not been triggered.

[0041] After completing field verification, the operating condition distance is calculated for each retained sample and the target cycle in the actual boundary condition sequence. This is used to filter healthy samples under operating conditions that are the same as or similar to the target cycle. The operating condition distance is calculated using the following formula: ; in, For the first The working condition distance between a healthy sample and the target cycle. , , and These are the distance weights corresponding to the inlet temperature of the cooling medium, the flow rate of the cooling medium, the fan speed, and the duration of the control distribution section, respectively. For the first The inlet temperature of the cooling medium for each healthy sample. The inlet temperature of the cooling medium for the target cycle. For the first Cooling medium flow rate of each healthy sample The cooling medium flow rate for the target cycle. For the first The fan speed of each healthy sample The fan speed for the target period. For the first Duration of the control allocation segment for each healthy sample The duration of the control allocation segment for the target cycle is determined. , , and These are the normalized baseline values ​​for the corresponding fields.

[0042] The reason for adopting this operating condition distance definition is that the reference power allocation generation model learns the control allocation law under a healthy state and specific heat dissipation boundary and duration conditions. If the operating condition deviation is too large, the sample will lose its demonstrative significance.

[0043] In this embodiment, only healthy samples with a working condition distance of no more than 0.15 are retained. For example, when the cooling medium inlet temperature of the target cycle is 30°C, the cooling medium flow rate is 11L / min, the fan speed is 3000r / min, and the duration of the control allocation segment is 1.5s, only healthy samples that meet the above threshold conditions are written into the healthy control allocation sample set, and the remaining samples are deleted.

[0044] After screening, each retained sample is organized into a sample structure consisting of an "operating condition status field plus a demonstration power allocation field." The operating condition status field includes at least the cooling medium inlet temperature, cooling medium flow rate, fan speed, and control allocation segment duration. The demonstration power allocation field represents the power allocation value for the corresponding period under healthy conditions. S210 outputs a healthy control allocation sample set for S220 to read directly.

[0045] S220. First, read the health control allocation sample set and the actual boundary condition sequence, and verify whether the operating condition status field and the demonstration power allocation field exist simultaneously in the health control allocation sample set. If a sample lacks the demonstration power allocation field, or if any key item in the operating condition status field is empty, then delete the sample.

[0046] After verification, a baseline power allocation generation model is established using imitation learning. Here, imitation learning means using the demonstration power allocation field in the health control allocation sample set as the learning target and the corresponding operating condition field as the input, so that the model learns the power assignment rules of the target embedded power module under different boundary conditions in a healthy state, rather than directly copying a fixed power value.

[0047] The loss value during model training is calculated using the following formula: ; in, For training loss values, This represents the number of samples used in the training process. For the first The model output power assignment value corresponding to each training sample. For the first The demonstration power assignment value corresponding to each training sample. The reason for using the absolute deviation average as the training loss is that this form of loss can directly constrain the deviation between the output power assignment value and the healthy demonstration value, avoiding excessive amplification of the training results by individual abnormal samples.

[0048] After training, the actual boundary condition sequence is input into the reference power allocation generation model periodically, and the reference power allocation value is generated periodically and sorted by period number to form the reference power allocation sequence.

[0049] To ensure the effectiveness of subsequent compensation calculations, output verification is performed on the generated results: if the baseline power allocation value generated in a certain period is negative, or the change range compared with the previous period exceeds the preset change limit, the model output of that period is deleted and replaced with the demonstration power allocation value corresponding to the healthy sample with the smallest working condition distance.

[0050] The reason for setting the output checksum substitution rule here is that the imitation learning model may exhibit unstable outputs at sample boundaries, and S230 subsequently requires a continuous and comparable baseline power allocation sequence. S220 outputs the baseline power allocation sequence for S230 and S310 to read directly.

[0051] Specifically, the baseline power allocation generation model is constructed using a multilayer perceptron (MLP) or a recurrent neural network (RNN). Its input layer dimension matches the number of features in the operating condition field. At least one hidden layer is used to fit the nonlinear mapping relationship between the operating condition and power allocation. During training, the Adam optimizer is used to optimize the training loss value using gradient descent until the model's performance on the validation set stabilizes.

[0052] S230. First, read the reference power allocation sequence and the actual power allocation sequence, and verify that the cycle numbers of the two correspond one-to-one. If a certain cycle appears in only one sequence, delete that cycle, because compensation calculation cannot be performed when the corresponding value is missing. For the retained cycles, calculate the load relinquishment compensation value cycle by cycle. The calculation formula is: ; in, For the first The load compensation value for each cycle, For the first The baseline power allocation value for each cycle. For the first The actual power allocation value for each cycle.

[0053] The direction definition of "baseline power allocation value minus actual power allocation value" is adopted because this scheme needs to identify the reduction between the power that the target embedded power module should have undertaken under healthy control allocation conditions and the power it actually undertaken under actual control compensation conditions.

[0054] when When, it indicates that there is a load shedding phenomenon during that cycle; when When, it indicates that there is no load shedding phenomenon in that cycle; when When this occurs, it indicates that there is a reverse load increase phenomenon in this cycle. At this time, the cycle is not deleted, but instead written into the compensation direction identifier field of the load-bearing compensation record, so that subsequent S410-S430 and S510-S530 can identify the change in compensation direction.

[0055] After the difference calculation is completed, a load-yielding compensation record will be generated for each cycle. Each load-yielding compensation record includes at least the cycle number, the reference power allocation value, the actual power allocation value, the load-yielding compensation value, the compensation direction identifier, and the validity identifier. Then, the load-yielding compensation sequence will be output by sorting the cycle numbers.

[0056] At this point, the load-yielding compensation sequence output by S230 will serve as the direct input for S410-S430 to calculate the compensation difference sequence and generate the pseudo-health judgment index sequence, and will also serve as the direct input for S510-S530 to identify the pseudo-health trend segment and calculate the degree of load-yielding compensation enhancement. Meanwhile, the reference power allocation sequence output by S220 will serve as the direct input for S310-S330 to construct the equal load reference thermal cycle simulation input set, thus ensuring that all outputs from this segment can continue to be used in subsequent steps.

[0057] S3 specifically includes the following sub-steps: S310. First, read the reference power allocation sequence, the actual boundary condition sequence, and the fixed structure parameter set corresponding to the target embedded power module, and verify whether the source identifiers of the three are consistent. Here, the source identifier refers to the module identifier, the simulation condition identifier, and the version identifier, which are used to ensure that the reference power allocation sequence, the actual boundary condition sequence, and the fixed structure parameter set read all correspond to the same target embedded power module and the same structure version.

[0058] If any of the three source identifiers is inconsistent, the data set is deleted and not included in the reference thermal cycle simulation construction process. This is because inconsistent sources could mistakenly write parameters from other modules or other versions of the structure into the current simulation. After verification, the fixed structure parameter set is categorized by type.

[0059] The fixed structural parameter set consists of a subset of static parameters and a subset of current state parameters. The subset of static parameters describes the geometric and material properties of the target embedded power module that remain unchanged at the design level, including at least chip length, chip width, package layer thickness, copper layer thickness, insulating layer thickness, material thermal conductivity, and basic parameters of heat dissipation boundaries. The subset of current state parameters describes the structural states that have been formed in the current identification stage and will not be replaced in this round of reference thermal cycle simulation, including at least interface state parameters, initial values ​​of contact thermal resistance, and interlayer connection state indicators.

[0060] The reason why the above parameters are uniformly classified into the fixed structural parameter set is that this scheme needs to construct "the reference thermal cycle response after switching the actual power distribution to the reference power distribution under the condition that the current structural state remains unchanged", so that the subsequent difference can specifically characterize the thermal response masking effect caused by control compensation, without introducing the additional variable of structural change.

[0061] After the parameters are locked, continue to verify whether the reference power allocation sequence and the actual boundary condition sequence correspond one-to-one according to the period number; if a certain period only has a reference power allocation value but no actual boundary condition record, or only has an actual boundary condition record but no reference power allocation value, then delete that period.

[0062] For the retention period, the reference power allocation value, cooling medium inlet temperature, cooling medium flow rate, fan speed, control allocation segment duration, and fixed structural parameter set reference identifier are written into the same cycle-by-cycle reference simulation input record, and sorted by cycle number to form an equal load reference thermal cycle simulation input set.

[0063] It should be noted that the cooling medium inlet temperature, cooling medium flow rate, fan speed, and duration of the control distribution segment in the actual boundary condition sequence are used to define the boundary values ​​that change periodically, while the basic parameters of the heat dissipation boundary in the fixed structural parameter set are used to define the heat dissipation connection relationship and boundary type. The two are not redundant.

[0064] For example, when the reference power allocation value for the 12th cycle is 31.5kW, the cooling medium inlet temperature is 29℃, the cooling medium flow rate is 10.8L / min, the fan speed is 2950r / min, and the duration of the control allocation segment is 1.4s, the above fields, together with the fixed structure parameter set reference identifier, constitute a cycle-by-cycle reference simulation input record for the 12th cycle. The S310 outputs the equal load reference thermal cycle simulation input set for direct reading by the S320.

[0065] S320. First, read the equal load reference thermal cycle simulation input set and verify whether each cycle-by-cycle reference simulation input record contains the cycle number, reference power allocation value, cooling medium inlet temperature, cooling medium flow rate, fan speed, control allocation segment duration, and fixed structure parameter set reference identifier. If any key field is missing, delete the corresponding record.

[0066] For the retained records, the bat algorithm is used to search for reference thermal cycle simulation parameters. The reference thermal cycle simulation parameters include at least the time step allocation factor, boundary response matching coefficient, and cyclic load transition parameters.

[0067] The time step allocation factor is used to correct the discrete time step of each cycle in the simulation solution process to improve the solution accuracy of the thermal response peak and fall-off segments; the boundary response matching coefficient is used to adjust the coupling strength between the cycle-by-cycle boundary values ​​and the basic parameters of the heat dissipation boundary; the cycle load transition parameter is used to control the smoothness of the power input transition from the previous cycle to the next cycle between adjacent cycles.

[0068] To make the search target of the bat algorithm clear, the comprehensive fitness value is calculated for each set of candidate reference thermal cycle simulation parameters. The calculation formula is as follows: ; in, The overall fitness value of the candidate reference thermal cycle simulation parameter set. Boundary condition matching error is used to characterize the deviation between the boundary response in the reference thermal cycle simulation and the actual boundary condition sequence. The periodic energy input consistency error is used to characterize the deviation between the integral value of the power input in each cycle and the corresponding energy of the reference power allocation value in the reference thermal cycle simulation. This is to account for the continuity error of the simulation curve, used to characterize the degree of temperature abrupt change at the junction of thermal responses between adjacent cycles. , and These are the weighting coefficients for the corresponding error terms.

[0069] The reason for adopting this comprehensive fitness definition is that this scheme does not pursue the minimum of any single error, but simultaneously requires consistent boundary responses, consistent energy inputs, and continuous curve connections. Only in this way can the simulation results of the reference thermal cycle be reliably compared with the characteristic sequences of subsequent actual thermal cycles. In the iterative process of the Bat Algorithm, an initial candidate parameter set is first randomly generated based on preset upper and lower limits, and then the candidate parameter set is updated round by round; where, the... The correction time step for each cycle satisfies: ; in, For the first The corrected time step for each cycle in the reference thermal cycle simulation. For the first The original time step of each cycle, For the first The time step allocation factor corresponds to each cycle. If a candidate parameter set causes the corrected time step to exceed the preset upper limit, or causes the boundary response matching coefficient or the cycle load transition parameter to exceed each custom domain, then the candidate parameter set is deleted.

[0070] Furthermore, if the trial calculation results corresponding to a certain candidate parameter set show non-physical temperature jumps, fewer reference cycles than input cycles, single-cycle energy input deviations exceeding preset upper limits, or temperature differences between adjacent cycle connection points exceeding preset thresholds, then the candidate parameter set is deleted and will not participate in subsequent iterations.

[0071] After continuous iterations, when the decrease in the overall fitness value does not exceed 0.5% in 5 consecutive iterations, the search is terminated, and the candidate parameter set with the smallest current overall fitness value is output as the optimized reference thermal cycle simulation parameter set.

[0072] For example, when a set of candidate parameters reduces the boundary condition matching error, periodic energy input consistency error, and simulation curve continuity error to 0.03, 0.05, and 0.02, respectively, this set of parameters can be used as the preferred output. The S320 outputs the optimized set of reference thermal cycle simulation parameters for direct reading by the S330.

[0073] S330: First, read the optimized reference thermal cycle simulation parameter set and the equal load reference thermal cycle simulation input set, then execute the reference thermal cycle simulation to obtain the periodic temperature response results of the target embedded power module under the reference power distribution conditions.

[0074] Then verify whether the simulation results for each cycle simultaneously include the highest temperature of the cycle, the lowest temperature of the cycle, the starting temperature of the heating segment, and the starting time of the heating segment; if any key result is missing, delete the corresponding cycle.

[0075] For the retained cycle, the same feature extraction rules as S130 are used to generate the equal load reference thermal cycle feature sequence. That is, the temperature cycle amplitude is extracted by the difference between the highest temperature and the lowest temperature of the cycle, the change in thermal resistance of the cycle is extracted by the difference in equivalent thermal resistance between adjacent cycles, and the temperature rise slope of the cycle is extracted by the ratio of the temperature change to the time change between the starting temperature of the heating segment and the highest temperature of the cycle.

[0076] After feature extraction, the obtained periodic features are sorted by period number, and the equal-load reference thermal cycle feature sequence is output. Here, only the valid corresponding periods in the reference thermal cycle simulation results are retained, and the actual thermal cycle feature sequence is not modified in this step; the uniform alignment, deletion of missing periods, and retention of valid periods between the actual thermal cycle feature sequence and the equal-load reference thermal cycle feature sequence are performed when constructing the comparison sample set in S410.

[0077] At this point, the equal load reference thermal cycle characteristic sequence output by S330 will be used together with the actual thermal cycle characteristic sequence and the load-bearing compensation sequence in subsequent steps S410-S430 to generate the compensation difference sequence and the pseudo-health judgment index sequence, thereby ensuring that the output of this section can continue to be used in subsequent steps and form a complete closed loop.

[0078] S4 specifically includes the following sub-steps: S410. First, read the actual thermal cycle characteristic sequence, the equal load reference thermal cycle characteristic sequence, and the load-yielding compensation sequence, and verify whether the module identifier, cycle number range, and validity identifier of the three are consistent. If the module identifier is inconsistent, delete all corresponding records. The reason is that there is no direct comparison relationship between the thermal response and compensation behavior of different target embedded power modules. If the cycle number range is inconsistent, only the cycle number range that the three have in common is retained. If any object of a certain cycle is missing from the three, delete all corresponding records of that cycle from the three. The reason for adopting the rule of deleting three for missing one is that the subsequent compensation difference calculation and pseudo-health judgment must be based on the premise that the three input objects of the same cycle exist at the same time. Retaining the missing cycle will generate invalid output.

[0079] After completing the first round of verification and deletion, the actual temperature cycle amplitude, reference temperature cycle amplitude, actual cycle thermal resistance change, reference cycle thermal resistance change, actual cycle temperature rise slope, reference cycle temperature rise slope, load compensation value, compensation direction identifier, and validity identifier within the same cycle are written into the same cycle-by-cycle comparison sample record according to the cycle number, and sorted by cycle number to form a comparison sample set.

[0080] The periodic comparison sample records here are not simply spliced ​​together, but are the only input records for subsequent difference calculations, so each record retains complete fields.

[0081] For example, when the 18th cycle simultaneously contains an actual temperature cycle amplitude of 14.2℃, a reference temperature cycle amplitude of 18.6℃, an actual cycle temperature rise slope of 6.1℃ per second, a reference cycle temperature rise slope of 7.4℃ per second, and a load compensation value of 3.8kW, the above fields together constitute a cycle-by-cycle comparison sample record for the 18th cycle. S410 outputs the comparison sample set for S420 to read directly.

[0082] S420. First, read the comparison sample set and verify whether each periodically comparison sample record contains the actual feature field, the reference feature field, and the load compensation field simultaneously; if a record is missing any field, delete the record.

[0083] For the retained records, the compensation difference is calculated uniformly in the direction of "reference value minus actual value". The reason is that this scheme needs to identify the difference between the thermal response that the target embedded power module should exhibit under the equal load reference condition and the thermal response it exhibits under the actual controlled operation condition. When the difference is positive, it indicates that the actual thermal response of the cycle is less than the equal load reference thermal response, which means that the cycle may be masked by control compensation.

[0084] No. The differences in temperature cycle amplitude, thermal resistance change, and temperature rise slope over each cycle are calculated using the following formulas: ; in, For the first Temperature cycle amplitude difference over one cycle, For the first The reference temperature cycle amplitude for each cycle, For the first The actual temperature cycle amplitude for each cycle.

[0085] ; in, For the first The difference in thermal resistance change over each cycle. For the first The change in thermal resistance over a reference period of one cycle. For the first The actual change in thermal resistance over each cycle.

[0086] ; in, For the first The difference in the temperature rise slope over each cycle. For the first The reference cycle temperature rise slope for each cycle. For the first The actual temperature rise slope of each cycle. After the difference calculation is completed, the three types of differences are normalized according to the corresponding reference characteristic values ​​to obtain the normalized temperature cycle amplitude difference, the normalized periodic thermal resistance change difference, and the normalized periodic temperature rise slope difference. The reason for using normalization is that the three types of characteristics have different dimensions and cannot be directly involved in the subsequent index fusion without unified processing.

[0087] For each cycle, the normalized three types of differences, the load compensation value, the compensation direction identifier, and the validity identifier are written into the same cycle-by-cycle compensation difference record, and then sorted by cycle number to form a compensation difference sequence.

[0088] If the reference temperature cycle amplitude is 18.6℃ and the actual temperature cycle amplitude is 14.2℃ in the 18th cycle, then the temperature cycle amplitude difference for that cycle is 4.4℃. If the corresponding load compensation value is positive, then that cycle is marked as a positive masking candidate cycle in the compensation difference sequence. S420 outputs the compensation difference sequence for S430 to read directly.

[0089] The normalization process specifically employs the Min-Max Normalization method, mapping the original differences to... Range. For the load compensation value. Its normalization benchmark is taken from the rated power value of the target embedded power module, thereby eliminating the impact of different dimensions on the fusion index. The impact.

[0090] S430. First, read the compensation difference sequence and verify whether each cycle-by-cycle compensation difference record contains the normalized temperature cycle amplitude difference, the normalized cycle thermal resistance change difference, the normalized cycle temperature rise slope difference, and the load-bearing compensation value. If any key field is missing, delete the record.

[0091] For retained records, pseudo-health assessment indicators are generated according to unified fusion rules. The pseudo-health assessment indicators for each cycle are calculated using the following formula: ; in, For the first Pseudo-health indicators for each cycle, For the first Normalized temperature cycle amplitude difference over each cycle For the first The difference in the normalized change of thermal resistance over each period For the first The difference in the normalized temperature rise slope of each cycle. For the first The normalized load compensation value after each cycle , , and These are the weight coefficients for the corresponding items.

[0092] The reason for adopting this fusion method is that the pseudo-health state is not triggered by a single thermal characteristic difference, but is formed by the thermal response difference and the load-yielding compensation behavior. After completing the calculation of the single-cycle index, a sliding judgment window of three consecutive cycles is used to perform trend judgment on the pseudo-health judgment index sequence. If the difference in normalized temperature cycle amplitude, the difference in normalized cycle thermal resistance change, and the difference in normalized cycle temperature rise slope of at least two cycles within the window maintain a positive relationship with the normalized load-yielding compensation value, then the window is judged as a direction-consistent window. If the pseudo-health judgment index of the subsequent cycle within the same window is greater than that of the previous cycle, then the difference between the indexes of the two adjacent cycles is defined as the expansion amplitude. If the expansion amplitude is continuously positive, it indicates that the control compensation masking effect is increasing.

[0093] Furthermore, if the absolute value of the rate of change corresponding to the actual thermal cycle characteristic sequence within a certain sliding judgment window does not exceed the preset stable threshold or decreases continuously, while the rate of change corresponding to the equal load reference thermal cycle characteristic sequence increases continuously and the load compensation value increases continuously, then the pseudo health judgment index for each period within the window is incrementally corrected and written into the pseudo health judgment index sequence.

[0094] For example, when the pseudo-health assessment indexes for three consecutive cycles are 0.42, 0.57, and 0.71, respectively, and the corresponding load compensation values ​​are 2.9kW, 3.4kW, and 3.8kW, respectively, it indicates that the target embedded power module has formed a continuously enhancing positive masking trend within this window.

[0095] Finally, the pseudo-health assessment indicators corresponding to all valid periods are sorted by period number, and a pseudo-health assessment indicator sequence is output. Thus, the pseudo-health assessment indicator sequence output in S430 will continue to be used in subsequent S510-S530 to form pseudo-health trend segments, determine staged degradation assessment results, and generate degradation trend identification results, thereby ensuring that the output of this segment is continued to be used in subsequent steps and forms a strict closed loop.

[0096] S5 specifically includes the following sub-steps: S510. First, read the pseudo-health judgment index sequence, actual thermal cycle characteristic sequence, equal load reference thermal cycle characteristic sequence, and load-yielding compensation sequence, and verify whether the module identifier, cycle number range, and validity identifier of the four are consistent. If the module identifier is inconsistent, delete the corresponding record. If the cycle number range is inconsistent, only retain the cycle number range that is covered by the four. If any object of a certain cycle is missing in the four, delete all corresponding records of that cycle in the four. The reason is that the pseudo-health trend segment must be based on the simultaneous existence of the judgment index, actual thermal response, reference thermal response, and load-yielding compensation value of the same cycle.

[0097] After verification, three consecutive periods are used as the initial trend identification window, and joint trend analysis is performed on each window. Here, the pseudo-health judgment index sequence is defined as the primary criterion to identify whether the masking effect continues to increase; the actual thermal cycle characteristic sequence and the isoload reference thermal cycle characteristic sequence are defined as auxiliary criteria to confirm whether the enhancement simultaneously meets the condition of "actual thermal response is stable or decreasing, and reference thermal response is increasing"; the load-bearing compensation sequence is defined as the compensation intensity criterion to confirm whether the control compensation continues to increase.

[0098] When the pseudo-health judgment index continuously increases within a certain window, and the absolute value of the change rate of the temperature cycle amplitude and the periodic temperature rise slope corresponding to the actual thermal cycle characteristic sequence does not exceed the preset stable threshold or continuously decreases, while the temperature cycle amplitude and the periodic temperature rise slope corresponding to the equal load reference thermal cycle characteristic sequence continuously increase, and the load compensation value continuously increases, the window is determined as the trend start window.

[0099] For subsequent cycles after the trend initiation window, if the pseudo-health judgment indicator continues to increase, or if there is only one cycle of decline and the decline does not exceed 5% of the previous cycle's indicator value, and the load compensation value does not show a continuous decline, then the subsequent cycle is merged into the current pseudo-health trend segment; if the above conditions are not met for two consecutive cycles, then the current pseudo-health trend segment is terminated.

[0100] After completing the segment boundary extraction, calculate the duration period, the initial pseudo-health judgment index within the segment, the final pseudo-health judgment index within the segment, the average pseudo-health judgment index within the segment, the initial load-bearing compensation value within the segment, and the final load-bearing compensation value within the segment for each pseudo-health trend segment, and write the above fields into the segment-by-segment trend record. The growth rate of the indicator for each pseudo-healthy trend segment is calculated using the following formula: ; in, For the first The growth rate of indicators in the pseudo-health trend segment. For the first The pseudo-health assessment indicators corresponding to the end of each pseudo-health trend segment. For the first The pseudo-health judgment indicators corresponding to the starting period of each pseudo-health trend segment. For the first The number of durations of pseudo-healthy trend segments.

[0101] The reason for using the average increment between the end-of-cycle indicator and the beginning-of-cycle indicator to represent the indicator growth rate is that this quantity can directly reflect the growth intensity during a continuous masking process, avoiding the amplification of results by a single abnormal cycle. Furthermore, the... The degree of load compensation enhancement for the pseudo-health trend segment is calculated using the following formula: ; in, For the first The degree of compensation for the false health trend segment is enhanced. For the first The load compensation value corresponding to the end of the pseudo-health trend segment. For the first The load compensation value corresponding to the starting period of the pseudo-health trend segment. For the first The number of durations of pseudo-healthy trend segments.

[0102] The reason for adopting this definition is that this scheme does not only consider the increase in thermal response difference, but also simultaneously confirms whether the control compensation continues to increase. For example, if a pseudo-health trend segment extends from the 21st cycle to the 25th cycle, with an initial pseudo-health judgment index of 0.46 and an ending pseudo-health judgment index of 0.81, an initial load-bearing compensation value of 2.8kW, and an ending load-bearing compensation value of 4.0kW, then the growth rate of the index and the degree of load-bearing compensation enhancement for this segment are both positive, indicating that this segment meets the condition for continuous enhancement. S510 outputs the pseudo-health trend segment for S520 to read directly.

[0103] S520. First, read the pseudo-health trend segment and verify whether each segment trend record contains the start period number, end period number, number of duration periods, indicator growth rate, and degree of load compensation enhancement. If any key field is missing, delete the corresponding trend record. If the number of duration periods is less than the preset minimum number of periods of 3, delete the trend record. The reason is that when the number of duration periods is insufficient, it is impossible to distinguish between short-term fluctuations and continuous degradation.

[0104] For retaining trend records, first determine the number of duration periods for each segment. growth rate of indicators and the degree of compensation for the load being transferred Perform normalization based on the extreme values ​​of historical trend samples to obtain the corresponding normalized statistics. and Subsequently, a unified stage scoring rule is used to generate staged degradation judgment results. The stage score for each pseudo-health trend segment is calculated using the following formula: ; in, For the first The stage score of the pseudo-health trend segment. This represents the normalized number of duration periods. The normalized growth rate of the indicator. To enhance the normalized load compensation level, , and These are the weight coefficients for the corresponding items.

[0105] The reason for adopting this stage scoring rule is that the degree of pseudo-health degradation is not determined by a single moment's judgment indicator, but by the duration, intensity of growth, and intensity of compensatory enhancement.

[0106] After the stage score is calculated, the stage score is compared with the preset grading threshold. When the stage score is less than the first threshold, the trend segment is judged as a mild occlusion degradation stage. When the stage score is not less than the first threshold and less than the second threshold, the trend segment is judged as a moderate occlusion degradation stage. When the stage score is not less than the second threshold, the trend segment is judged as a severe occlusion degradation stage.

[0107] If a trend record contains invalid cycles and the number of invalid cycles exceeds 1, the trend record will be deleted and no staged degradation judgment result will be generated.

[0108] After the classification is completed, the module identifier, start cycle number, end cycle number, number of duration cycles, stage score, stage level, and trend validity identifier are written into the same staged degradation judgment record, and sorted by the start cycle number to form the staged degradation judgment result. S520 outputs the staged degradation judgment result for S530 to read directly.

[0109] S530. First, read the phased degradation judgment results and verify whether each phased degradation judgment record contains the module identifier, start cycle number, end cycle number, number of duration cycles, phase level, and trend validity identifier. If any key field is missing, delete the corresponding record.

[0110] For the retained records, degradation trend identification results are generated according to the stage level. The degradation trend identification results are output in the form of result records. Each degradation trend identification result record includes at least the module identifier, the pseudo-health status identifier of the transferred vehicle, the start period number of the pseudo-health trend segment, the end period number of the pseudo-health trend segment, the number of duration periods, the stage level, risk warning information, and the result generation timestamp.

[0111] The pseudo-health status identifier of the transferred vehicle is generated according to the following rules: when there is at least one valid staged degradation judgment record with a stage level not lower than the mild masking degradation stage, the pseudo-health status identifier of the target embedded power module is written as valid; when there is no staged degradation judgment record that meets the conditions, the status identifier is written as invalid.

[0112] The risk warning information is not free text, but standardized fields generated by mapping according to the stage level; among them, the risk warning information corresponding to the mild coverage degradation stage is "continuous monitoring", the risk warning information corresponding to the moderate coverage degradation stage is "arrange maintenance and verification", and the risk warning information corresponding to the severe coverage degradation stage is "trigger priority maintenance".

[0113] For example, if the target embedded power module is in the moderate masking degradation stage corresponding to the 21st to 25th cycles, the risk warning information in the degradation trend identification result record corresponding to the module is written as "arrange maintenance and verification".

[0114] Finally, all valid degradation trend identification results are recorded and sorted by result generation timestamp and module identifier, and the degradation trend identification results are output. At this point, S530 transforms the aforementioned cycle-by-cycle thermal response difference, load-bearing compensation behavior, and pseudo-health judgment index sequence into directly callable staged identification results, thereby completing the closed loop of degradation trend identification of the target embedded power module thermal cycling simulation results.

[0115] In this embodiment, the weighting coefficients involved in each formula (such as...) (etc.) is determined through a prior knowledge base combined with sensitivity analysis. Specifically, by changing a single input feature (such as temperature cycle amplitude) and observing its contribution to the judgment result of known degraded samples, the contribution is normalized and used as the initial weight of the corresponding item, and then fine-tuned according to the error feedback under actual working conditions.

[0116] All the above formulas are performed using dimensionless numerical calculations; the relevant formulas are based on empirical models that approximate the real situation, obtained through extensive data collection and software simulation fitting. The preset parameters and thresholds involved in the formulas can be conventionally set and adjusted by those skilled in the art according to the physical constraints of the actual application scenario.

[0117] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0118] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0119] In conclusion, the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying degradation trends in embedded power module thermal cycle simulation results, characterized by, Includes the following steps: S1. Obtain the control allocation record, sensor sampling record and thermal cycling simulation result of the target embedded power module, perform time alignment, period slicing and module mapping on the record, and generate the actual power allocation sequence, actual boundary condition sequence and actual thermal cycling characteristic sequence including temperature cycle amplitude, period thermal resistance change and period temperature rise slope. S2. Based on the actual boundary condition sequence, select health control allocation samples from the healthy operation sample library, use imitation learning to generate the reference power allocation sequence of the target embedded power module in the healthy state, and combine it with the actual power allocation sequence to generate the load-yielding compensation sequence. S3. Under the condition that the fixed structural parameter set remains unchanged, construct the equal load reference thermal cycle simulation input set based on the reference power distribution sequence and the actual boundary condition sequence, optimize the reference thermal cycle simulation parameters using the bat algorithm, execute the reference thermal cycle simulation, and generate the equal load reference thermal cycle feature sequence. S4. Construct a comparison sample set based on the actual thermal cycle characteristic sequence, the equal load reference thermal cycle characteristic sequence, and the load-bearing compensation sequence, and generate a compensation difference sequence and a pseudo-health judgment index sequence.

2. The method of claim 1, wherein, Also includes: S5. Based on the pseudo-health judgment index sequence, actual thermal cycle characteristic sequence, equal load reference thermal cycle characteristic sequence and load-bearing compensation sequence, extract pseudo-health trend segments, generate staged degradation judgment results, and output the degradation trend identification results of the target embedded power module.

3. The method for identifying degradation trends in embedded power module thermal cycling simulation results according to claim 1, characterized in that, S1 specifically includes: Read the controller log stream, smart sensor sampling stream, and thermal cycling simulation result file corresponding to the target embedded power module to form the original running dataset; Perform time alignment, cycle slicing and module mapping on the original running dataset, delete data segments with missing timestamps, abnormal self-check status, mismatched operating condition identifiers and non-closed cycles, and generate a set of cycle running data blocks; Based on the set of periodic operation data blocks, extract actual power distribution information, actual boundary condition information and periodic thermal response information to generate actual power distribution sequence, actual boundary condition sequence and actual thermal cycle characteristic sequence.

4. The method for identifying degradation trends in embedded power module thermal cycling simulation results according to claim 1, characterized in that, S2 specifically includes: Read the healthy operation sample library and the actual boundary condition sequence, verify the integrity of the sample fields, and filter healthy samples according to the operating condition distance between the cooling medium inlet temperature, cooling medium flow rate, fan speed and the duration of the control allocation segment and the target cycle, and generate a healthy control allocation sample set; Based on the health control allocation sample set and the actual boundary condition sequence, a benchmark power allocation generation model is established using imitation learning, and the benchmark power allocation sequence is output.

5. The method for identifying the degradation trend of embedded power module thermal cycling simulation results according to claim 4, characterized in that, Also includes: The reference power allocation sequence and the actual power allocation sequence are matched by period number. The load waiver compensation value for each period is calculated and sorted by period number to generate the load waiver compensation sequence.

6. The method for identifying degradation trends from thermal cycling simulation results of an embedded power module according to claim 1, characterized in that, S3 specifically includes: Read the reference power distribution sequence, the actual boundary condition sequence, and the fixed structural parameter set. Under the condition that the fixed structural parameter set remains unchanged, construct the equal load reference thermal cycle simulation input set. Based on the input set of the equal load reference thermal cycle simulation, the bat algorithm is used to search and optimize the reference thermal cycle simulation parameters, and the optimized reference thermal cycle simulation parameter set is output. The reference thermal cycle simulation is performed based on the optimized reference thermal cycle simulation parameter set, and the equal load reference thermal cycle feature sequence is generated according to the same feature extraction rules as the actual thermal cycle feature sequence.

7. The method for identifying degradation trends from thermal cycling simulation results of an embedded power module according to claim 1, characterized in that, S4 specifically includes: Read the actual thermal cycle characteristic sequence, the equal load reference thermal cycle characteristic sequence, and the load-yielding compensation sequence; verify the consistency of module identifier, cycle number, and validity identifier; delete the corresponding cycle record that is missing for any object; and construct a comparison sample set according to the same cycle number. Based on the comparison sample set, the differences in temperature cycle amplitude, periodic thermal resistance change, and periodic temperature rise slope between the reference value and the actual value are calculated respectively, and a compensation difference sequence is generated after normalization.

8. The method for identifying the degradation trend of embedded power module thermal cycling simulation results according to claim 7, characterized in that, Also includes: Based on the compensation difference sequence and the load-bearing compensation value, a pseudo-health judgment index sequence sorted by period is generated.

9. The method for identifying the degradation trend of embedded power module thermal cycling simulation results according to claim 2, characterized in that, S5 specifically includes: Read the pseudo-health judgment index sequence, actual thermal cycle characteristic sequence, equal load reference thermal cycle characteristic sequence and load-yielding compensation sequence, use a sliding window to extract pseudo-health trend segments, and calculate the number of duration periods, index growth rate and load-yielding compensation enhancement degree of each pseudo-health trend segment. The classification is based on the number of consecutive cycles, the growth rate of indicators, and the degree of load compensation enhancement for each pseudo-health trend segment, generating a staged degradation judgment result.

10. The method for identifying the degradation trend of embedded power module thermal cycling simulation results according to claim 9, characterized in that, Also includes: Based on the phased degradation judgment results, a degradation trend identification result is generated, which includes module identifier, pseudo-health status identifier of the load, corresponding period interval, phase level and risk warning information.

Images