An adaptive wide-temperature-range power supply controller system and control method

Abstract

This invention proposes an adaptive wide-temperature-range power supply controller system and control method, belonging to the field of power supply controller technology. The system includes a multi-dimensional temperature sensing module, an adaptive FNN-PID control module, a dead-time dynamic optimization module, an active thermal stress management module, an edge digital twin health management module, an impedance environment sensing module, an intelligent protection state machine module, and a line voltage drop compensation module. The modules interact with each other through a shared RAM area of ​​a dual-core MCU, employing a double-buffering mechanism to ensure cross-core data consistency. The system defines a global state vector as the core data structure shared by all modules. Compared with existing technologies, the technical solution of this invention achieves a leap from passive defense to active adaptive and predictive control.

Description

Technical Field

[0001] This invention belongs to the field of power controller technology, and particularly relates to an adaptive wide temperature range power controller system and control method. Background Technology

[0002] The power controller mentioned in this invention is specifically used in the auxiliary power system of the combiner box controller of photovoltaic power plants in the Shago desert region.

[0003] In such real-world operating environments, the equipment faces extremely harsh climatic and physical challenges: Extremely wide temperature range: the system needs It maintains stable and reliable operation within a wide ambient temperature range of 40°C to +85°C. Core components within the system (such as dual-core microcontrollers) need to be covered... Operating temperature range of 40°C to +125°C, while wide-bandgap silicon carbide (SiC) MOSFET switching devices in the main power stage need to withstand temperatures as low as... Extreme junction temperature fluctuations of 55°C and up to +175°C.

[0004] Complex physical environment erosion: The desert and Gobi areas are characterized by intense sandstorm erosion, dry dust accumulation, and condensation caused by diurnal temperature variations. These complex external physical conditions pose a serious threat to the insulation performance, heat dissipation efficiency, and long-term structural reliability of power supply systems.

[0005] For the aforementioned extreme application scenarios, conventional power controllers and existing basic implementation solutions have exposed many technical bottlenecks, which constitute the microscopic technical background for the key breakthrough of this invention: 1. The control parameters lack wide-range dynamic adaptive capability. Existing power converters (such as conventional Buck topology converters) typically employ fixed control parameters tuned at a nominal temperature of 25°C. However, over a wide temperature range, the physical characteristics of power devices drift significantly (e.g., the on-state voltage drop and on-resistance of SiC MOSFETs change markedly with temperature). Current technologies rely on fixed dead times (e.g., a nominal 75ns) and static PID control parameters, which are ineffective at extremely low temperatures (e.g., low temperatures). At 40°C or under extremely hot conditions, it can easily lead to a sharp increase in device switching losses, deterioration of output voltage ripple, or insufficient closed-loop phase margin of the system.

[0006] 2. Lack of proactive control mechanisms to cope with extreme thermal stress Traditional power supply thermal management often relies on passive heat sink design or simple single high-temperature threshold derating / shutdown protection. In extremely cold startup scenarios, existing equipment lacks an active preheating mechanism, often making it difficult to establish output smoothly; while in summer high-temperature full-load conditions, existing technologies lack sophisticated means of reducing thermal stress.

[0007] 3. Gaps in life prediction and multi-failure mode health monitoring capabilities Conventional systems typically only provide macroscopic hardware-level overvoltage, overcurrent, or overtemperature protection. However, under long-term cyclic thermo-mechanical stress, power modules can develop various hidden competing failure modes, such as solder layer fatigue, bond wire delamination, and package delamination. Existing solutions struggle to run reduced-order thermal models in real-time on computationally limited edge devices, making it impossible to estimate the dynamic degradation of junction-to-case or case-to-environment equivalent thermal resistance online. This prevents the system from integrating multiple individual damage levels to output a comprehensive health index (HI), thus hindering accurate prediction of remaining useful life (RUL) and support for proactive maintenance decisions.

[0008] 4. Lack of perception and collaborative self-recovery mechanisms for complex polluted environments. In the face of sandstorms and subsequent condensation conditions, existing power supply systems typically lack online sensing capabilities for environmental impedance characteristics, making it impossible to accurately identify whether equipment is clean, dry with accumulated dust, or damp and contaminated. Therefore, existing solutions cannot automatically intercept high-voltage startup to prevent insulation breakdown in damp conditions, nor can they use impedance spectrum scanning to link piezoelectric self-cleaning or low-power circulating dehumidification functions, making them highly susceptible to hard failures in harsh environments. Summary of the Invention

[0009] To address the above problems, this invention proposes an adaptive wide-temperature-range power supply controller system and control method.

[0010] The adaptive wide-temperature-range power controller system includes a multi-dimensional temperature sensing module, an adaptive FNN-PID control module, a dead-time dynamic optimization module, an active thermal stress management module, an edge digital twin health management module, an impedance environment sensing module, an intelligent protection state machine module, and a line voltage drop compensation module. The modules interact with each other through the shared RAM area of ​​the dual-core MCU, and a double buffering mechanism is used to ensure cross-core data consistency. The system defines a global state vector as the core data structure shared by all modules. The multi-dimensional temperature sensing module is used to output the system temperature state vector; The adaptive FNN-PID control module is used to output control quantities to the power stage drive circuit; The dead time dynamic optimization module is used to output the optimal dead time to the PWM generator dead time register; The active thermal stress management module is used to execute active thermal control strategies; The edge digital twin health management module is used for predictive maintenance and intelligent health management; The impedance environment sensing module is used to perform impedance spectrum measurement, environmental status identification, and self-cleaning and dehumidification linkage control by utilizing power topology multiplexing. The intelligent protection state machine module is used to integrate information from all modules to manage the system's operating status and make decisions on protection actions; The line voltage drop compensation module is used to calculate the voltage drop compensation amount of the remote cable in real time based on the ambient temperature and load current, and adjust the voltage control reference.

[0011] The system allocates two equal-length storage areas in the shared RAM for writing and reading data from the global state vector.

[0012] The global state vector includes a temperature state sub-vector, a control state sub-vector, a health state sub-vector, an environmental state sub-vector, and a system mode sub-vector. The temperature state sub-vector is updated by the multi-dimensional temperature sensing module in each high-frequency real-time control cycle; The control state subvector is updated by the adaptive FNN-PID control module and the dead-time dynamic optimization module in each intermediate frequency adaptive adjustment cycle; The health status sub-vector is updated by the edge digital twin health management module in low-frequency health management and environmental perception. The environmental state sub-vector is updated by the impedance environment sensing module after the impedance spectrum scan is completed; The system mode subvector is updated by the intelligent protection state machine module in each intermediate frequency adaptive adjustment cycle.

[0013] The multi-dimensional temperature sensing module outputs a temperature state vector to the shared RAM area through physical temperature acquisition, virtual junction temperature estimation, online ESR estimation, and multi-source information fusion.

[0014] The multidimensional temperature sensing model acquires physical temperature through four thermistor sensors deployed on the controller PCB.

[0015] The multidimensional temperature sensing model uses the on-state voltage drop signal of the power stage sampling circuit to estimate the virtual junction temperature.

[0016] The multidimensional temperature sensing model is based on the voltage AC ripple component and inductor current ripple component collected by the system in real time, and obtains the online ESR estimate by calculating the ripple peak ratio.

[0017] The multi-source information fusion includes at least the fusion calculation of physical temperature acquisition values ​​and virtual junction temperature estimation values.

[0018] The adaptive FNN-PID control module is responsible for adaptive adjustment of PID parameters based on temperature conditions and calculation of PWM duty cycle, so as to output control quantity to the power stage drive circuit.

[0019] The adaptive FNN-PID control module performs adaptive adjustment of fuzzy neural network PID parameters and calculation of PWM duty cycle based on the virtual junction temperature estimate and ESR online estimate in the temperature state sub-vector, the current state of the intelligent protection state machine module in the system mode sub-vector, and the working mode of the active thermal stress management module.

[0020] The dead time dynamic optimization module is responsible for calculating the optimal dead time based on the estimated virtual junction temperature in the temperature state sub-vector, and outputting it to the PWM generator dead time register.

[0021] The active thermal stress management module executes two active thermal management strategies: nighttime insulation and high-temperature frequency modulation peak shaving, based on the virtual junction temperature estimate in the temperature state sub-vector, the current state of the intelligent protection state machine module in the system mode sub-vector, and the request flag set by the adaptive FNN-PID control module.

[0022] The edge digital twin health management module performs predictive maintenance and intelligent health management based on the virtual junction temperature estimate in the temperature state sub-vector and the power calculation mode flag in the active thermal stress management module.

[0023] The predictive maintenance and intelligent health management includes real-time simulation of reduced-order thermal models, operation of digital twin simulation engine, online correction of model parameters, rainflow counting and cumulative damage calculation, multi-failure mode damage fusion, prediction of health indicator degradation trends, prediction of remaining life, and predictive maintenance decision support.

[0024] The impedance environment sensing module performs a complete impedance spectrum scan and updates the environment state subvector when the system is in standby or the load rate is below 10%. When the system is in normal operation and the load rate is consistently above 10%, the narrowband fast impedance detection mode is enabled.

[0025] The intelligent protection state machine module manages the system's operating status and makes decisions on protection actions based on the following information: Temperature information in the temperature state subvector; The cumulative damage level, thermal resistance information, and comprehensive health index in the health status subvector; The environment state in the environment state subvector; The current control parameters in the control state subvector.

[0026] The management of the system's operating status and the decision-making for protection actions include: Determine and execute job state transitions; Dust accumulation adaptive dynamic derating; Recovery from abnormal working conditions.

[0027] The line voltage drop compensation module calculates the voltage drop compensation amount of the remote cable in real time based on the temperature information in the temperature state sub-vector and the load current obtained by the power level sampling circuit.

[0028] This patent also proposes a control method for an adaptive wide-temperature-range power controller. The method is used to control the adaptive wide-temperature-range power controller system and includes system initialization, high-frequency real-time control, medium-frequency adaptive adjustment, low-frequency health management, and environmental perception.

[0029] The system initialization includes: Perform a self-test on the system hardware; Physical temperature is acquired using a thermistor sensor; The impedance environment sensing module performs rapid impedance self-test to quickly determine the environment category. The edge digital twin health management module is initialized to verify whether the deviation between the initial output of the digital twin and the physical measurement is within a reasonable threshold range; Make an initial working mode decision to determine the working mode of the module.

[0030] The initial operating mode decision, to determine the operating mode of the module, includes: If the ambient temperature is lower than the set temperature or the environment is damp and dirty, the intelligent protection state machine module enters the preheating state; if the ambient temperature is not lower than the set temperature and the environment is not damp and dirty, the intelligent protection state machine module enters the soft start state; if the comprehensive health index loaded by the edge digital twin health management module is greater than the set value, then regardless of the temperature and environmental conditions, the intelligent protection state machine module enters the dynamic derating state to protect the severely degraded power module.

[0031] The high-frequency real-time control includes the controller system cyclically executing the following high-frequency control functions after the intelligent protection state machine module enters the soft-start state or normal operation state: Synchronous data sampling: to collect data such as voltage, current, and on-state voltage drop of the controller; Virtual junction temperature estimation update: The junction temperature is calculated by substituting the on-state voltage drop sample value into the calibration equation, and then the virtual junction temperature estimation value is updated after 8-point moving average filtering. ESR Online Estimation Update: Based on the voltage AC ripple component and the inductor current ripple component, calculate and update the online ESR estimate; Observer iteration for multi-source information fusion: Perform a one-step iteration of the discrete state equations of the multi-source information fusion observer; PID control law calculation: including voltage error calculation, where the reference voltage is the value after compensation by the line voltage drop compensation module; Hardware protection monitoring: If the inductor current or output voltage exceeds the rated threshold range, the hardware comparator immediately triggers an interrupt and forces the intelligent protection state machine module to switch to the emergency cut-off state.

[0032] The intermediate frequency adaptive adjustment includes the controller system cyclically performing the following intermediate frequency adaptive adjustment: Fuzzy inference based on temperature data is used to weighted synthesize PID adjustment coefficients; The final PID parameters are synthesized through a neural network, while backpropagation is performed to update the network weights. Mode arbitration and execution of the active thermal stress management module: The active thermal stress management module reads the request flag and the status of the intelligent protection state machine module, determines whether to accept the request according to the arbitration priority, and executes the corresponding mode switch; PID parameters are adjusted in conjunction with frequency modulation: If the current mode is frequency modulation mode of the active thermal stress management module, the PID parameters are scaled twice to maintain the phase margin at a value not lower than the set phase margin value. Dead time update: Read the estimated virtual junction temperature from the temperature state subvector, calculate and update the temperature-compensated dead time; Digital twin iteration: Perform a one-step iteration of the reduced-order state equation; Compensation calculation of line voltage drop compensation module: Calculate the voltage drop compensation amount of the far-end cable, and update and adjust the reference voltage after upper limit clamping for high frequency cycle use; Intelligent protection state machine module state transition judgment: Check whether the comprehensive health indicators exceed the warning threshold; if so, trigger the transition to dynamic depreciation state. Double-buffered pointer flip: Reverses the read / write roles of two equal-length memory regions through atomic operations.

[0033] The low-frequency health management and environmental perception include the controller system performing the following low-frequency health management and environmental perception in a non-preemptive manner: Perform thermistor sensor sampling and thermal field reconstruction: Perform filtered sampling on the thermistor sensor, and substitute it into the cubic spline interpolation algorithm to calculate the 8×6 grid temperature distribution map; Perform thermal resistance trend analysis: calculate the dust accumulation degradation coefficient. If the change in the dust accumulation degradation coefficient exceeds the set dust accumulation degradation coefficient, dynamically adjust the derating start temperature and notify the intelligent protection state machine module. Perform rainflow counting and damage update: Perform rainflow counting on the junction temperature sequence of the most recent 10 seconds, extract the thermal cycle, calculate the theoretical lifetime, and update the cumulative damage. Perform multi-failure mode damage fusion: After completing the damage update in the previous step, calculate the bond line lift-off damage and package delamination damage, and then weight and fuse the bond line lift-off damage, package delamination damage, and solder layer fatigue damage to obtain a comprehensive health index. Hourly degradation trend fitting and remaining useful life (RUL) prediction: Read the historical sequence of the daily average value of the comprehensive health index for the past 30 days, use the weighted least squares method to fit the linear degradation trend, and extrapolate the nominal value, optimistic value and pessimistic value of remaining useful life based on the degradation slope. Impedance spectrum measurement of the impedance environment sensing module; Online model calibration of the edge digital twin health management module: Compare the digital twin output with the observer output every hour or after the operating condition changes. If the deviation exceeds the set deviation value for 10 consecutive times, start the recursive least squares incremental correction. Perform health log persistence every hour: Write a snapshot of the current health status to non-volatile storage.

[0034] The method also includes state transition linkage control, which is an event-driven process triggered by the intelligent protection state machine module within the medium-frequency adaptive adjustment cycle.

[0035] The aforementioned state transition linkage control includes the linkage process for the intelligent protection state machine module to transition from the normal operation state to the dynamic derating state, as well as the linkage process for recovery from the emergency disconnection state.

[0036] Compared with the prior art, the technical solution of the present invention has the following advantages: The core advantage of this invention lies in its leap from passive defense to proactive adaptive and predictive control. Through the deep integration of multi-dimensional temperature sensing and adaptive FNN-PID control, this invention can dynamically adjust control law parameters and dead time in real time based on extreme temperature fluctuations and the drift of the physical characteristics of silicon carbide power devices across the entire temperature range. More significantly, this invention pioneers an active thermal stress management mechanism, utilizing an innovative main power stage topology to achieve controllable self-heating of the low duty cycle inductor circulating current during extremely cold nights and peak clipping through switching frequency modulation during extremely hot days. This fundamentally overcomes the technical pain points of traditional equipment, such as difficulty in starting up in extremely cold weather and high-temperature thermal stress concentration.

[0037] Furthermore, this invention offers unparalleled advantages over existing technologies in terms of long-term reliability and adaptability to complex environments. It embeds a digital twin health management engine within the edge-side microcontroller, enabling real-time simulation and fusion of multiple competing failure modes, such as solder layer fatigue, bond wire delamination, and package delamination, to achieve precise quantification of comprehensive health indicators and prediction of remaining service life (RUL). Simultaneously, this invention cleverly reuses power topology to construct an impedance environment sensing module, eliminating the need for expensive external environmental sensors and accurately identifying the system's dry, dusty, or damp state online. Upon detecting severe contamination or condensation, the system automatically locks the high-voltage startup process and activates hardware-linked piezoelectric ceramic self-cleaning or inductor circulating evaporation dehumidification mechanisms, completely avoiding the fatal insulation breakdown risk caused by physical erosion from sand and dust and condensation. This significantly improves the maintenance-free cycle and high reliability of the power system in extreme natural environments. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of an adaptive wide-temperature-range power controller system framework according to an embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram illustrating the control method of an adaptive wide-temperature-range power controller according to an embodiment of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0042] It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components, units, circuits, and steps described in these embodiments do not limit the scope of the invention.

[0043] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0044] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0045] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0046] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0047] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0048] In terms of hardware topology, this embodiment provides an adaptive wide-temperature-range power controller for the extreme environment of the Gobi Desert. The main power stage of this controller adopts a two-phase interleaved parallel Buck topology, with each phase containing one upper transistor and one freewheeling diode (or a synchronous rectification lower transistor). The two phases operate with a 180° phase interleaving to reduce output current ripple. The power switching device is a wide-bandgap silicon carbide (SiC) MOSFET, specifically the Wolfspeed C3M0065100K, with a rated voltage of 1000V, a rated continuous drain current of 35A, an on-resistance of 65mΩ at 25°C, and an allowable operating junction temperature range of [missing information]. 55°C to +175°C.

[0049] The output filtering stage employs a hybrid parallel configuration of solid-state polymer aluminum electrolytic capacitors and multilayer ceramic capacitors (MLCCs). Specifically, it uses four 150μF / 63V solid-state polymer electrolytic capacitors connected in parallel with two 22μF / 100V X7R MLCC ceramic capacitors. The solid-state polymer capacitors provide the main energy storage and low-frequency filtering; their ESR (Equivalent Series Resistance) temperature characteristics are superior to traditional liquid electrolytic capacitors but still vary with temperature. The MLCCs provide high-frequency decoupling and switching ripple bypass, complementing each other to cover filtering needs across the entire frequency range from DC to tens of MHz. The output filter inductor uses a Kool Mu 60μF iron-silicon-aluminum core, with an inductance of 47μH and a saturation current of 30A.

[0050] The core controller uses a TI TMS320F28388D dual-core microcontroller, which includes a C28x real-time digital signal processor core (200MHz, used for high-speed PWM (Pulse Width Modulation) control, observer operation, fuzzy neural network inference, and PID (Proportional-Integral-Derivative) parameter adaptive calculation) and an ARM Cortex-M4 application core (125MHz, used for digital twin calculation, impedance spectrum analysis, and low-frequency health management tasks). It has a built-in 16-bit high-precision ADC (analog-to-digital converter) and a wide operating temperature range. The system operates from 40°C to +125°C. It also integrates a Lattice iCE40 FPGA coprocessor for high-speed ADC sampling synchronous triggering, cubic spline interpolation, and fine-tuning of the PWM (Pulse Width Modulation) dead zone.

[0051] The system's rated parameters are: input voltage range 200Vdc to 400Vdc, output voltage 48Vdc, rated output power 1kW, and nominal switching frequency 100kHz. This converter is used as an auxiliary power supply for the combiner box controller of a photovoltaic power station in the Shagohuang area.

[0052] Based on the hardware topology, the power controller in this embodiment of the invention is functionally divided into eight core modules. This embodiment explicitly incorporates line voltage drop compensation functionality into the module architecture, ensuring a one-to-one correspondence between module declarations and actual functional units. These eight modules are as follows: Module 1: Multidimensional temperature sensing module, responsible for physical temperature acquisition, virtual junction temperature estimation, online ESR (Equivalent Series Resistance) estimation and multi-source information fusion, outputting the system temperature state vector.

[0053] Module 2: Adaptive FNN-PID control module (Fuzzy Neural Network; Proportional-Integral-Derivative), responsible for adaptive adjustment of fuzzy neural network PID (Proportional-Integral-Derivative) parameters based on temperature conditions and calculation of PWM (Pulse Width Modulation) duty cycle, outputting control quantities to the power stage drive circuit.

[0054] Module 3: Dead Time Dynamic Optimization Module (DTDO) is responsible for dynamically calculating the optimal dead time based on the junction temperature of the power devices and outputting it to the dead time register of the PWM (Pulse Width Modulation) generator.

[0055] Module 4: Active Thermal Stress Management (ATSM) module, responsible for implementing two active thermal management strategies: nighttime insulation and high-temperature frequency modulation peak shaving.

[0056] Module 5: Edge Digital Twin Health Management Module (DTHM) is responsible for real-time simulation of the reduced-order thermal model, operation of the digital twin inference engine, online correction of model parameters, rainflow counting and cumulative damage calculation, multi-failure mode damage fusion, prediction of health indicator degradation trends, prediction of remaining life, and predictive maintenance decision support.

[0057] Module 6: Impedance Environment Sensing Module (IBEP), responsible for impedance spectrum measurement, environmental condition identification, and self-cleaning and dehumidification linkage control using power topology multiplexing.

[0058] Module 7: Intelligent Protection State Machine (FSM) module, responsible for integrating information from all modules to manage the system's operating status and make decisions on protection actions.

[0059] Module 8: Line Voltage Drop Compensation Module (LDC), responsible for calculating the voltage drop compensation amount of the remote cable in real time based on ambient temperature and load current, and adjusting the voltage control reference.

[0060] Based on the above, this invention proposes an adaptive wide-temperature-range power supply controller system, such as... Figure 1 As shown, the system includes a multi-dimensional temperature sensing module, an adaptive FNN-PID control module (Fuzzy Neural Network; Proportional-Integral-Derivative), a dead-time dynamic optimization module, an active thermal stress management module, an edge digital twin health management module, an impedance environment sensing module, an intelligent protection state machine module, and a line voltage drop compensation module. The above modules interact with each other through the shared RAM (Random Access Memory) area of ​​the dual-core MCU (Microcontroller Unit), and a double buffering mechanism is used to ensure cross-core data consistency. The system defines a global state vector as the core data structure shared by all modules. The multi-dimensional temperature sensing module is used to output the system temperature state vector; The adaptive FNN-PID control module (Fuzzy Neural Network; Proportional-Integral-Derivative) is used to output control quantities to the power stage drive circuit. The dead time dynamic optimization module is used to output the optimal dead time to the PWM (Pulse Width Modulation) generator dead time register; The active thermal stress management module is used to execute active thermal control strategies; The edge digital twin health management module is used for predictive maintenance and intelligent health management; The impedance environment sensing module is used to perform impedance spectrum measurement, environmental status identification, and self-cleaning and dehumidification linkage control by utilizing power topology multiplexing. The intelligent protection state machine module is used to integrate information from all modules to manage the system's operating status and make decisions on protection actions; The line voltage drop compensation module is used to calculate the voltage drop compensation amount of the remote cable in real time based on the ambient temperature and load current, and adjust the voltage control reference.

[0061] The eight modules mentioned above exchange data through the shared RAM (RandomAccess Memory) area of ​​the dual-core MCU (Microcontroller Unit), employing a double-buffering mechanism to ensure cross-core data consistency. The system defines a global state vector G(k) as the core data structure shared by all modules, where k is the discrete time step and the global sampling period is 1ms. G(k) contains five sub-vectors: Temperature state subvector (k) contains six elements: ambient temperature Power stage case temperature Temperature of capacitor region Temperature of magnetic components Virtual junction temperature estimate Online estimates of ESR (Equivalent Series Resistance) This subvector is updated by module 1 every high-frequency real-time control cycle (10 μs). The virtual junction temperature estimate is... Online estimates of ESR (Equivalent Series Resistance) Updated by module 1 in each high-frequency real-time control cycle (10μs); ambient temperature Power stage case temperature Temperature of capacitor area Temperature of magnetic components The sample is taken by a 10Hz thermistor and refreshed after thermal field reconstruction in the low-frequency cycle, while the previous refresh value is maintained in the high-frequency cycle.

[0062] Control state subvector (k) contains six elements: the current PID (Proportional-Integral-Derivative) three parameters (Current proportional gain coefficient) (Current integral gain coefficient) (Current differential gain coefficient), current PWM (Pulse Width Modulation) duty cycle Current switching frequency and current dead zone time This subvector is updated by modules 2 and 3 in each intermediate frequency adaptive adjustment period (1ms).

[0063] Health state subvector (k) contains seven elements: current cumulative damage. Equivalent thermal resistance from junction to shell Equivalent thermal resistance from radiator to environment Dust accumulation degradation coefficient Estimated remaining useful life Comprehensive health index (HI) and slope of degeneration trend This sub-vector is updated by module 5 in low-frequency health management and environmental perception.

[0064] Environment state subvector (k) contains four elements: environmental category identifier (0 for clean, 1 for dry dust, and 2 for damp and dirty), low-frequency impedance modulus, high-frequency impedance imaginary part, and the timestamp of the last complete scan. This subvector is updated by module 6 after the impedance spectrum scan is completed.

[0065] System mode subvector (k) contains five elements: the current state of the Intelligent Protection State Machine (FSM) module (values ​​S0 to S6), the operating mode of the Active Thermal Stress Management (ATSM) module (atsm_mode, value 0: Normal, 1: Insulation, 2: Frequency Modulation), the configuration update flag, the impedance environment sensing module (IBEP) scan request flag, and the alarm flag. This sub-vector is updated by module 7 in each intermediate frequency adaptive adjustment cycle.

[0066] Meanwhile, to address the data consistency issue when the C28x real-time core and the Cortex-M4 application core asynchronously access the global state vector G(k), this embodiment of the invention allocates two equal-length storage regions for G(k) in shared RAM (Random Access Memory), labeled Buffer_A and Buffer_B respectively, with each buffer occupying approximately 256 bytes of storage space. At any given time, the real-time core, acting as a producer, writes the latest data to one of the buffers, while the application core, acting as a consumer, reads data from the other buffer.

[0067] After each round of intermediate frequency (IF) task computation by the real-time core, a buffer role flip is performed via an atomic flag bit in the Inter-Processor Communication (IPC) register: the currently write buffer is marked as readable, and the currently read buffer is marked as writable. The flip operation itself is a single 32-bit register write, taking less than 10 ns to execute, with no intermediate states. The flip frequency is equal to the IF task frequency of 1 kHz. This mechanism ensures that the application core reads a complete and consistent state snapshot at any given time, rather than a partially updated intermediate state.

[0068] Furthermore, in this embodiment of the invention, the adaptive wide-temperature-range power controller system allocates two equal-length storage areas in the shared RAM (RandomAccess Memory) for writing and reading data.

[0069] The global state vector includes a temperature state sub-vector, a control state sub-vector, a health state sub-vector, an environmental state sub-vector, and a system mode sub-vector. The temperature state sub-vector is updated by the multi-dimensional temperature sensing module in each high-frequency real-time control cycle. The control state sub-vector is updated by the adaptive FNN-PID control module and the dead-time dynamic optimization module in each mid-frequency adaptive adjustment cycle. The health state sub-vector is updated by the edge digital twin health management module in low-frequency health management and environmental sensing. The environmental state sub-vector is updated by the impedance environment sensing module after impedance spectrum scanning is completed. The system mode sub-vector is updated by the intelligent protection state machine module in each mid-frequency adaptive adjustment cycle.

[0070] Module 1 (Multidimensional Temperature Sensing Module) receives analog signals from four NTC (Negative Temperature Coefficient) thermistor sensors, as well as the on-state voltage drop signal and output voltage ripple signal from the power stage sampling circuit. After fusion calculation by the Luenberger state observer, the temperature state sub-vector is written to the current write buffer. This sub-vector is read and used by modules 2, 3, 4, 5, and 8.

[0071] Furthermore, the multi-dimensional temperature sensing module outputs the system temperature state sub-vector to the shared RAM (Random Access Memory) area through physical temperature acquisition, virtual junction temperature estimation, online ESR (Equivalent Series Resistance) estimation, and multi-source information fusion.

[0072] The multidimensional temperature sensing model acquires physical temperature through four thermistor sensors deployed on the controller PCB (Printed Circuit Board).

[0073] The multidimensional temperature sensing model uses the on-state voltage drop signal of the power stage sampling circuit to estimate the virtual junction temperature.

[0074] The multidimensional temperature sensing model is based on the voltage AC ripple component and inductor current ripple component collected by the system in real time, and obtains the online estimated value of ESR (Equivalent Series Resistance) by calculating the ripple peak ratio.

[0075] The multi-source information fusion includes at least the fusion calculation of physical temperature acquisition values ​​and virtual junction temperature estimation values.

[0076] In the distributed physical sensor network of the multi-dimensional temperature sensing module, this embodiment deploys four NTC (Negative Temperature Coefficient) thermistors on the controller PCB (Printed Circuit Board). The selected thermistors are Murata NCP18XH103F03RB type 100kΩ±25°CNTC, with a B value of 3380K and a measurement range of [missing information]. The operating temperature range is 40°C to +125°C, with an accuracy of ±1°C. The four sensors are arranged as follows: The first sensor is located at the chassis air inlet or the heat dissipation channel of the outer casing to collect the ambient reference temperature. It uses a glass-encapsulated NTC (Negative Temperature Coefficient) to resist dust corrosion. The second sensor is close to the heat sink of the main power MOSFET module and uses a chip-type NTC (Negative Temperature Coefficient) directly soldered to the copper plane to reduce the contact thermal resistance to below 0.5°C / W. The third sensor is located on the copper foil of the PCB (Printed Circuit Board) near the output electrolytic capacitor bank. The fourth sensor is located on the surface of the inductor core. Each sensor is digitally acquired through a 16-bit ADC (Analog-to-Digital Converter) built into the MCU (Microcontroller Unit) at a sampling rate of 10Hz.

[0077] In this embodiment of the invention, the PCB (Printed Circuit Board) is divided into an 8×6 rectangular grid. The nth sensor is located at a grid coordinate, and its measured temperature is the corresponding value. Cubic spline functions are constructed along the x and y directions respectively, satisfying the interpolation conditions at each sensor location and possessing continuous first and second derivatives. For any grid point, the temperature estimate is obtained through bidirectional weighted interpolation, with the weights normalized based on the distance from that point to the nearest sensor. This calculation is performed by the FPGA (Field Programmable Gate Array) within each low-frequency task cycle (1Hz), and the calculation results are stored in a circular buffer for remote operation and maintenance queries.

[0078] For virtual junction temperature estimation, the MOSFET forward voltage drop is used as a temperature-sensitive electrical parameter. During the turn-on phase of each PWM (Pulse Width Modulation) switching cycle, within a steady-state window approximately 300ns to 500ns after the turn-on instant and 200ns before the turn-off instant, a high-speed ADC (Analog-to-Digital Converter) triggered by the FPGA (Field Programmable Gate Array) synchronously samples the drain-source voltage. After the sampled values ​​are filtered through eight moving averages, they are substituted into the calibration equation to calculate the estimated virtual junction temperature.

[0079] in The on-state voltage drop is calibrated at 25°C. For the actual measured drain-source on-state voltage drop, α is a temperature coefficient (typically 5mV / °C for the selected SiC MOSFET). For reference temperature, This is an estimated value for the virtual junction temperature.

[0080] For online ESR (Equivalent Series Resistance) estimation, the system acquires the AC ripple components of the output voltage and the ripple components of the inductor current in real time. The estimated ESR is calculated using the ripple peak ratio, where the peak output voltage ripple is extracted using a bandpass filter (center frequency 100kHz, bandwidth ±10%). This estimation is performed by the real-time core in each high-frequency control cycle. When the estimated ESR exceeds three times the nominal value at 25°C, it is marked as an extremely low temperature condition or capacitor aging condition.

[0081]

[0082] In the above formula, This is an online estimate of the equivalent series resistance (ESR). The peak value of the output voltage ripple. This represents the peak value of the inductor current ripple.

[0083] In terms of multi-source information fusion, a Luenberger state observer is used to fuse complementary information from physical sensors (high accuracy but bandwidth of approximately 1 Hz) and virtual observers (bandwidth of the same order as the switching frequency but affected by model accuracy). The continuous domain state equation of the observer is:

[0084] in To estimate the temperature state vector (6-dimensional, corresponding to six key thermal nodes, including estimated virtual junction temperatures), A is the thermal network state matrix, and B is the power input matrix. For real-time calculation of power loss at each node, Let C be the physical sensor measurement value, C be the output matrix, and L be the observer gain matrix. To estimate the error, Let L be the rate of change of the temperature state vector. The design of L places the observer poles 3 to 5 times to the left of the system poles. After discretization, it runs in a real-time kernel at 100 kHz, with a single computation taking approximately 0.8 μs.

[0085] The adaptive FNN-PID control module is responsible for adaptive adjustment of PID (Proportional-Integral-Derivative) parameters based on temperature conditions and calculation of PWM (Pulse Width Modulation) duty cycle to output control quantities to the power stage drive circuit.

[0086] The adaptive FNN-PID control module performs adaptive adjustment of fuzzy neural network PID (Proportional-Integral-Derivative) parameters and calculation of PWM (Pulse Width Modulation) duty cycle based on the virtual junction temperature estimate and ESR (Equivalent Series Resistance) online estimate in the temperature state sub-vector, the current state of the intelligent protection state machine module in the system mode sub-vector, and the working mode of the active thermal stress management module.

[0087] As described above, module 2 obtains the virtual junction temperature estimate and online ESR estimate from the temperature state sub-vector, as well as the current state of the intelligent protection state machine module and the operating mode of the active thermal stress management module from the system mode sub-vector. It then performs fuzzy inference and neural network forward computation, outputting updated PID (Proportional-Integral-Derivative) parameters and PWM (Pulse Width Modulation) duty cycle to the control state sub-vector. When the consequent of the fuzzy rule layer triggers a request from the active thermal stress management module, module 2 sets the corresponding request flag in the system mode sub-vector to request the initiation of insulation and frequency modulation peak clipping. This request flag is read and arbitrated by module 4 in the next intermediate frequency adaptive adjustment cycle.

[0088] Regarding the control architecture of the adaptive FNN-PID control module, the FNN-PID controller takes four variables as input: output voltage error. Error change rate Core temperature (Taking the 1-second moving average of the estimated junction temperature) and the temperature change rate dT / dt. The output is the PWM (Pulse Width Modulation) duty cycle adjustment Δd(n). This is the adjusted reference voltage after voltage drop compensation by module 8; The measured output voltage during the nth control cycle; The first-order difference of the error reflects how fast the error changes, corresponding to the derivative control term of PID (Proportional-Integral-Derivative), used to suppress overshoot; the actual duty cycle of PWM (Pulse Width Modulation) is d(n) = d(n 1)+Δd(n) is the sum of the duty cycle and duty cycle adjustment amount of the previous cycle.

[0089] Core temperature Mapped to five fuzzy linguistic variables, the membership functions are defined as follows: DeepFreeze uses a left trapezoidal function: when T c ≤ The membership degree is 1 at 35°C. 35°C <T c < The membership degree at 20°C is ( 20 T c ) / 15, when T c ≥ The membership degree is 0 at 20°C. The low-temperature (Cold) function uses a trigonometric function, with the center point... 15°C, left boundary 25°C, right boundary 5°C. Normal temperature uses a trigonometric function, center point 25°C, left boundary 5°C, right boundary 50°C. Hot temperature uses a trigonometric function, center point 65°C, left boundary 45°C, right boundary 80°C. Extreme temperature uses a right trapezoidal function: when T... c The membership degree is 0 at ≤75°C, and 0 at 75°C. <T c At <85°C, the membership degree is (T) c 75) / 10, when T c The membership degree is 1 at ≥85°C. The rate of temperature change dT / dt maps to three fuzzy linguistic variables: Falling (dT / dt < 2°C / min), Stable For 2°C / min≤dT / dt≤2°C / min and Rising (dT / dt>2°C / min), a trapezoidal membership function is used, with a transition zone width of ±0.5°C / min.

[0090] The fuzzy rule base contains 15 rules (5 temperature levels × 3 rate of change levels). Each rule's consequent is the adjustment coefficients for the three PID (Proportional-Integral-Derivative) parameters: ΔKp (proportional adjustment coefficient), ΔKi (integral adjustment coefficient), and ΔKd (derivative adjustment coefficient), as well as an optional ATSM request flag. The actual PID parameters are calculated through weighted synthesis.

[0091] Where μ i The activation strength of each rule, This is the adjustment factor for the proportional coefficient corresponding to the i-th rule. This is the baseline value for the proportionality coefficient. The actual proportional coefficient for the current control cycle is represented by Ki, and the integral and derivative coefficients of Ki and Kd are calculated in the same way.

[0092] Furthermore, a three-layer backpropagation (BP) neural network is cascaded after the fuzzy layer for long-term self-evolution. The input layer has four neurons (e, Δe, T). c , dT / dt), where e is the output voltage error in the input variables of the FNN-PID controller, Δe is the error change rate, and T c The core temperature is dT / dt, representing the rate of temperature change. The hidden layer has 8 neurons (using a sigmoid activation function), and the output layer has 3 neurons (Kp, Ki, and Kd are the final correction values, where Kp, Ki, and Kd are the proportional, integral, and derivative gain coefficients of the PID controller). The objective function is half the sum of squared errors. Stochastic gradient descent is used to update the weights online, with a learning rate η = 0.01, a momentum factor α = 0.9, and a maximum weight amplitude of ±5.0. Weight updates are performed every 1kHz mid-frequency task cycle.

[0093] The dead time dynamic optimization module is responsible for calculating the optimal dead time based on the estimated virtual junction temperature in the temperature state sub-vector and outputting it to the PWM (Pulse Width Modulation) generator dead time register.

[0094] Furthermore, the dead time dynamic optimization module reads the estimated virtual junction temperature from the temperature state subvector, calculates the temperature-compensated dead time, writes it into the control state subvector, and synchronously updates the dead time register of the PWM (Pulse Width Modulation) timer.

[0095] The formula for calculating dead time is:

[0096] in, The dynamic dead time at the current junction temperature. The nominal dead time is 75 ns, calibrated at 25°C. The compensation factor is 0.5 ns / °C. This is an estimated virtual junction temperature. =25°C is the reference temperature. This formula ensures that... The dead time is reduced to approximately 42.5 ns at 40°C and extended to approximately 125 ns at 125°C. The safety clamping range is 20 ns to 200 ns. The update frequency is 1 kHz.

[0097] The active thermal stress management module executes two active thermal management strategies: nighttime insulation and high-temperature frequency modulation peak shaving, based on the virtual junction temperature estimate in the temperature state sub-vector, the current state of the intelligent protection state machine module in the system mode sub-vector, and the request flag set by the adaptive FNN-PID control module.

[0098] Module 4 (ATSM) reads the estimated virtual junction temperature from the temperature state subvector, and reads the current state of the intelligent protection state machine module and the request flag set by the adaptive FNN-PID control module from the system mode subvector, performing mode determination and arbitration. The arbitration logic is as follows: if the FSM is currently in the S4 derating state, the ATSM ignores the PEAK_SHAVE_REQUEST flag (indicating a request to start frequency modulation peak clipping), because derating itself already limits power; if the FSM is currently in the S6 emergency cut-off state, the ATSM ignores all requests; otherwise, the ATSM accepts the request from Module 2 and evaluates whether to execute it. When switching operating modes, in addition to updating the atsm_mode field (value 0 for normal, 1 for heat preservation, 2 for frequency modulation) in the system mode subvector, the ATSM also sends the power calculation mode flag P_LOSS_MODE to Module 5, with a value of 0 for normal operation, 1 for heat preservation, and 2 for frequency modulation. This flag ensures that the digital twin model of Module 5 always uses power loss input that matches the current operating mode.

[0099] When the ATSM starts in heat preservation mode (atsm_mode=1), a low duty cycle inductor circulating current heating method is adopted. The system disconnects the main output from the load (by turning off the output isolation MOSFET) and operates the single-phase power stage in normal switching mode using a Buck topology with an extremely low duty cycle (0.5% to 5%). The high-side MOSFET is periodically turned on according to a controlled duty cycle, and the current is established through the output inductor; after the high-side MOSFET is turned off, the inductor current decays through the low-side MOSFET freewheeling current. The I²R loss generated by the inductor current in the MOSFET on-resistance and the inductor DC resistance achieves controllable and uniform heating of the power stage components. The switching frequency is set to 10kHz to reduce the proportion of switching losses. The heating power calculation formula is:

[0100] In the above formula, The heating power is the Joule heat power generated by the inductive circulating current across the circuit resistance. The effective value of the inductor current, the root mean square value of the inductor current in the circuit, determines the magnitude of the heating power; It is the on-resistance of the high-side MOSFET; It is an inductor and a DC resistor; It is the on-resistance of the low-side MOSFET. It is the total resistance of the circulating circuit.

[0101] Based on the parameters calculated in this embodiment, the on-resistance of each MOSFET in the two-phase power stage is... At 28°C, the resistance is approximately 37 mΩ (derived from the nominal value of 65 mΩ at 25°C, through the positive temperature coefficient of conduction resistance α). R (Converted), the DC resistance of the inductor is approximately 30mΩ, and the total resistance of the single-phase loop is... =37mΩ + 30mΩ + 37mΩ = 104mΩ. To obtain a heating power of 25W, the required effective value of the inductor current is... The current draw is approximately 15.5A, still far below the inductor saturation current of 30A and the MOSFET's rated continuous current of 35A, ensuring safe device operation. In discontinuous conduction mode, the peak inductor current is controlled within the range of 15A to 18A by adjusting the duty cycle (typically approximately 3% to 5%), corresponding to a continuously adjustable heating power between 20W and 30W. The target temperature for heat preservation is 20°C. Once the estimated virtual junction temperature reaches the target, the heating power is reduced to maintain the temperature within the range of 20°C ± 5°C. The heat preservation mode is exited when the ambient temperature rises above 10°C.

[0102] When ATSM starts and exits the heat preservation mode, it synchronously sets the P_LOSS_MODE flag (with a value of 0 for normal operation, 1 for heat preservation mode, and 2 for frequency modulation mode): it is set to 1 (heat preservation mode) upon startup and returns to 0 (normal mode) upon exit. After detecting a change in P_LOSS_MODE (0 for normal operation, 1 for heat preservation mode, and 2 for frequency modulation mode), module 5 (DTHM) switches the power loss input of the digital twin model to the I²R loss calculation formula corresponding to inductive circulating heating, ensuring that the model output is consistent with reality.

[0103] Regarding high-temperature frequency modulation peak clipping, when the ATSM starts frequency modulation mode (atsm_mode=2), the adjusted switching frequency is calculated as follows:

[0104] in It is the adjusted switching frequency; =100kHz is the nominal switching frequency; =0.4 is the frequency modulation coefficient; This is a virtual junction temperature estimate; =70°C is the frequency modulation start-up temperature; =120°C is the maximum temperature for frequency modulation. The lower frequency limit is 50kHz.

[0105] After frequency modulation is initiated, the ATSM synchronously performs the following actions: sets P_LOSS_MODE to 2 (indicating frequency modulation mode) to notify module 5 to switch modes; and writes the adjusted frequency into the control state sub-vector for module 2 to read. Upon detecting frequency adjustment, module 2 automatically scales the PID (Proportional-Integral-Derivative) parameters according to the following formula to maintain a phase margin of no less than 45°:

[0106] In the above formula, This is the adjusted proportional coefficient. This is the current (before adjustment) proportional coefficient. It is the adjusted integral coefficient. This is the current (before adjustment) integral coefficient. It is the adjusted switching frequency. =100kHz is the nominal switching frequency. When the estimated virtual junction temperature drops below 65°C, the ATSM exits frequency modulation mode and restores all flags and parameters.

[0107] The edge digital twin health management module performs predictive maintenance and intelligent health management based on the virtual junction temperature estimate in the temperature state sub-vector and the power calculation mode flag in the active thermal stress management module.

[0108] The predictive maintenance and intelligent health management includes real-time simulation of reduced-order thermal models, operation of digital twin simulation engine, online correction of model parameters, rainflow counting and cumulative damage calculation, multi-failure mode damage fusion, prediction of health indicator degradation trends, prediction of remaining life, and predictive maintenance decision support.

[0109] Module 5 (DTHM) reads the estimated virtual junction temperature from the temperature state subvector and the P_LOSS_MODE flag (0 for normal operation, 1 for insulation mode, and 2 for frequency modulation mode) from Module 4. It then performs a one-step state equation iteration of the Reduced Order Model (ROM). The digital twin junction temperature output by the model is compared with the junction temperature output by the observer, triggering online correction logic. Simultaneously, the rainflow counting submodule continuously processes the junction temperature time series and updates the cumulative damage degree in the health state subvector. Module 5 also performs multi-failure mode damage fusion calculation, weighting and fusing the independent damage degrees of solder layer fatigue, bond line lift-off, and package delamination into a comprehensive health index (HI). Based on the HI time series, Module 5 uses weighted least squares to fit the degradation trend and extrapolates to calculate the remaining useful life (RUL). When the recursive least squares (RLS) parameter correction is triggered, module 5 sets the internal correction flag CORRECTION_ACTIVE and records the correction time. Upon detecting this flag, the rainflow counting submodule automatically discards the temperature data for 5 seconds before and after the correction time, and only performs rainflow counting on the data outside the transition window. The update result of the health state subvector is read by module 7 for state transition decision-making.

[0110] The DTHM module is the core of this system for predictive maintenance and intelligent health management. Its functional architecture is divided into seven sub-modules: real-time simulation sub-module of reduced-order thermal model, sub-module of digital twin inference engine operation, sub-module of online correction of model parameters, sub-module of rainflow counting and cumulative damage calculation, sub-module of multi-failure mode damage fusion, sub-module of health indicator degradation trend prediction, and sub-module of remaining life prediction and predictive maintenance decision support.

[0111] In terms of real-time simulation of reduced-order thermal models, we have gone through the establishment of offline high-fidelity models, the process of reducing order through intrinsic orthogonal decomposition (POD), the derivation of the reduced-order state space matrix, discretization, and embedded deployment.

[0112] In establishing the offline high-fidelity model, a high-fidelity three-dimensional transient thermal model of the power module was first built offline using finite element thermal simulation software. This model comprises a complete multi-layer structure including a SiC wafer (350 μm thick, thermal conductivity 370 W / (m·K) @ 25°C), a solder layer (SAC305, 50 μm thick, thermal conductivity 58 W / (m·K)), a DBC aluminum nitride ceramic substrate (0.635 mm thick, thermal conductivity 170 W / (m·K)), a copper base plate (3 mm thick, thermal conductivity 385 W / (m·K)), and an aluminum finned heatsink. The model mesh has approximately 100,000 nodes, a time step of 1 μs, and spatial discretization using a tetrahedral unstructured mesh. Transient simulations were performed on this high-fidelity model under various operating conditions to generate a snapshot matrix. Specific operating conditions include: constant power step (4 levels: 50W, 100W, 150W, 200W), periodic power pulses (16 combinations with frequencies from 0.01Hz to 10Hz and duty cycles from 20% to 80%), linear power ramps (3 rates of ascent and descent), and random power sequences (4 typical daily power curves generated based on measured photovoltaic power output fluctuations in the desert). Simulations are performed with a 1ms step size, and the temperature field is downsampled according to the power module's thermal time constant (approximately 3-5s) (e.g., one frame stored every 0.2s). Each operating condition has approximately 9000 frames, totaling approximately 2.4 × 10⁻⁶ frames for the 27 operating conditions. 5 The frames form a snapshot matrix S(N×M, N≈100,000, M≈2.4×10). 5 The snapshot method is applied to the snapshot matrix S (for M×M S). S is decomposed and then projected back (or economical / truncated singular value decomposition), directly extracting the first r eigenorthogonal decomposition POD basis vectors, U is N×r, without the need to explicitly construct an N×N full matrix.

[0113] In the intrinsic orthogonal decomposition (POD) order reduction process, temperature snapshots of all operating conditions are arranged column-wise to form a snapshot matrix S with dimensions N×M, where N is the number of spatial nodes (approximately 100,000) and M is the total number of snapshots. Singular value decomposition (SVD) is then performed on the snapshot matrix.

[0114] In the above formula, S is the snapshot matrix, and U is an N×r vector matrix under the economical SVD, whose first r columns are the eigenorthogonal decomposition POD basis vectors. Let N×M be a singular value diagonal matrix. U is the transpose of a right singular vector matrix of dimension M×M. The first r columns of U are the eigenmodes (orthogonal decomposition) basis vectors (PODs), and the corresponding r maximal singular values ​​capture the main energy of the temperature field.

[0115] The modal truncation criterion uses the cumulative energy ratio criterion: selecting the smallest order r such that the sum of squares of the first r singular values ​​accounts for no less than 99.9% of the total sum of squares of singular values.

[0116] In the above formula is the i-th singular value (sorted in descending order); r is the truncation order (the smallest r to be selected); It is the sum of squares of the first r singular values, representing the total energy of the first r modes; It is the sum of squares of all singular values, representing the total energy of the temperature field; 0.999 is the energy retention ratio. For the power module thermal model of this embodiment, experiments show that r=20 to 30 can satisfy the above criteria. Take the first r columns of the POD basis matrix Φ as U, with dimension N×r.

[0117] The intrinsic orthogonal decomposition (POD) reduction process is a temperature field reduction method based on singular value decomposition (SVD). The core of it is to decompose high-dimensional temperature field data through intrinsic orthogonal decomposition (POD) to extract a few dominant modes. While retaining 99.9% of the energy, it significantly reduces the model dimensionality, providing a simplified model for subsequent rapid simulation and online estimation.

[0118] The derivation of the reduced-order state-space matrix involves the process of reducing the order of the power module's thermal model. The goal is to transform the high-fidelity finite element model into a low-dimensional state-space model through eigenorthogonal decomposition of the POD basis vectors, for use in real-time simulation and online estimation. The semi-discrete form of the high-fidelity finite element model is as follows:

[0119] Where M is the heat capacity matrix (N×N), K0 is the thermal conductivity matrix (N×N), F is the load distribution matrix, u1 is the power loss input vector, and T2 is the total node temperature vector. Let the temperature change rate vector be the vector. Project the temperature field onto the intrinsic orthogonal decomposition POD basis:

[0120] In the above formula, The eigenorthogonal decomposition is the POD basis matrix. Let be the reduced-order state vector, and r be the order of the model after reduction. Substituting into the original equation and left-multiplying by the transpose of Φ (Galerkin projection), we obtain the reduced-order state equation: 1 Where the reduced heat capacity matrix =Φ ·M·Φ(r×r), reduced-order thermal conductivity matrix =Φ ·K·Φ(r×r), reduced-order load matrix =Φ F,r represents the order (number of modes) of the model after the order reduction. This is the derivative of the reduced-order state vector, i.e., the rate of change of the lower-dimensional state variables. Transformed into standard state-space form:

[0121]

[0122] The system matrix = · Input matrix = · Output matrix The output matrix is ​​the one that maps from the reduced-order coordinates to the temperature of interest, and y is the output vector. It is formed by extracting the rows corresponding to the junction temperature nodes of the power device from the intrinsic orthogonal decomposition POD basis matrix Φ.

[0123] In terms of discretization and embedded deployment, the process involves converting the reduced-order thermal model of the continuous domain into a discrete form and deploying it to embedded platforms such as MCUs (Microcontroller Units). The reduced-order continuous-domain model is discretized using a zero-order hold with a sampling period Ts = 1ms.

[0124]

[0125] in, It is the reduced-order state vector at the k1th sampling time. It is the reduced-order state vector at the (k1+1)th sampling time. It is the input vector (power loss) at the k1th sampling time. It is the output vector at the k1th sampling time (focusing on the node temperature), and the discrete system matrix. =exp( · Discrete input matrix = ·( I)· , It is the output matrix (same as the continuous domain). It is the matrix of a continuous domain system (from a reduced-order model). The input matrix is ​​a continuous domain matrix (from a reduced-order model), exp( ) is the matrix exponentiation operation, and I is the identity matrix.

[0126] The discretized matrix is ​​stored as a 32-bit floating-point number in the MCU (Microcontroller Unit) flash memory. When r=25, It occupies 25 × 25 × 4 = 2500 bytes. and Each element occupies several hundred bytes, with a total storage size of approximately 5kB. Reduced-order state vector. It occupies 25×4=100 bytes of RAM. A single matrix-vector multiplication involves approximately 25×25=625 multiplication-addition operations, and the execution time on a Cortex-M4 core (125MHz) is approximately 100μs to 200μs.

[0127] The operation of the digital twin simulation engine includes initialization and cold start strategies, dynamic switching of power loss input, simulation result output, and deviation monitoring.

[0128] Regarding the initialization and cold start strategy, after the system is powered on, the digital twin inference engine needs to align the initial state of the reduced-order model with the physical system. The initialization strategy is as follows: the engine reads the initial sensor temperature value from the temperature state sub-vector and maps the physical temperature to the reduced-order state space using the pseudo-inverse of the intrinsically orthogonal decomposition (POD) basis matrix. The initial state is calculated through the pseudo-inverse mapping of the intrinsically orthogonal decomposition (POD) basis matrix: ,in The approximate reconstruction vector of the initial physical temperature field is obtained by expanding the measurements from four sensors to all nodes using cubic spline interpolation. Due to the orthogonality of the intrinsically orthogonal decomposition (POD) basis, the above pseudo-inverse simplifies to... .in, Let be the initial state vector of the reduced-order model. The eigenorthogonal decomposition is the POD basis matrix. This is the approximate reconstruction vector for the initial physical temperature field. After initialization, the initial deviation between the digital twin junction temperature output by the engine and the physical measurement value should not exceed ±2°C. Otherwise, the system records an initialization deviation alarm and prioritizes eliminating the deviation during the first recursive least squares (RLS) correction.

[0129] Regarding the dynamic switching of power loss input, the power loss input vector of the simulation engine dynamically switches the calculation method according to the P_LOSS_MODE flag (0 represents normal operation mode, 1 represents heat preservation mode, and 2 represents frequency modulation mode) passed by module 4. When P_LOSS_MODE=0 (normal operation mode), the power loss is calculated according to the standard formula:

[0130]

[0131]

[0132] In the above formula, It is the conduction loss of the MOSFET. This is the effective value of the load current. It is the on-resistance that varies with the estimated virtual junction temperature. The function, It is the switching loss of the MOSFET. It is the device's turn-off voltage. It refers to the turn-on and turn-off times of the device. It is the current switching frequency. This represents the total power loss of the switching transistor. The relationship between the on-resistance and temperature is as follows:

[0133] in =0.008 / °C is the positive temperature coefficient of the on-resistance of SiC MOSFET. = 65mΩ, which is the nominal on-resistance at 25℃. T3 is the estimated value of the current virtual junction temperature. It is the on-resistance at temperature T3.

[0134] When P_LOSS_MODE=1 (heat preservation mode), the power loss is calculated based on low duty cycle inductor circulating current heating:

[0135] In the above formula, It is the effective value of the inductor current. It is the on-resistance of the high-side MOSFET. It is an inductor and a DC resistor. It is the on-resistance of the low-side MOSFET. It is the power loss input vector, which serves as the input for the thermal model iteration.

[0136] When P_LOSS_MODE=2 (frequency modulation mode), the switching loss component is recalculated by replacing the nominal frequency with the actual adjustment frequency, while the conduction loss remains unchanged.

[0137] Regarding the output of simulation results and deviation monitoring, the junction temperature output, deviation monitoring, and data storage mechanism of the digital twin reduced-order thermal model are described. The simulation engine outputs the digital twin junction temperature in each intermediate frequency cycle (1ms):

[0138] Simultaneously calculate the deviation between the digital twin junction temperature and the observer junction temperature:

[0139] In the above formula, It is the digital twin junction temperature of the k3rd cycle. It is the output matrix (a matrix that maps from the reduced-order state to the temperature of the node of interest (junction temperature)). It is the reduced-order state vector of the k3rd period. Junction temperature deviation in the k3rd period This is a virtual junction temperature estimate output by a physical observer (such as the Luenberger observer). The deviation is stored in a circular buffer of length 100, which the correction submodule uses to determine whether to trigger a recursive least squares (RLS) update. Additionally, the simulation engine maintains a 72-hour junction temperature history curve buffer (downsampled at 10-second intervals, totaling 25,920 data points, occupying approximately 100kB of RAM) for use by the rainflow counting submodule and the degradation trend prediction submodule.

[0140] The online calibration of model parameters includes recursive least squares (RLS) incremental calibration and coupling protection between calibration and rainflow counting. The calibration trigger condition is one of the following two: timed triggering, i.e., the interval since the last calibration exceeds 1 hour; event triggering, i.e., the absolute value of the deviation exceeds 3°C for 10 consecutive sampling points (i.e., 10 consecutive ms) in the deviation ring buffer. The purpose of setting a continuity threshold for event triggering is to avoid false triggering of calibration by transient disturbances (such as short-term deviation spikes caused by load steps, ATSM mode switching, etc.).

[0141] In terms of incremental correction using recursive least squares (RLS), the goal of RLS correction is to incrementally correct the elements related to thermal resistance in the reduced-order state space matrix, so that the digital twin output approximates the physical measurement value. Specifically, the diagonal elements related to junction-to-case thermal resistance and case-to-heat sink thermal resistance in the reduced-order form of the thermal conductivity matrix are taken as the parameter vector θ to be identified, with dimension p (in this embodiment, p=4, corresponding to the two equivalent thermal resistances of junction-to-case and case-to-heat sink for each of the two-phase power modules).

[0142] The recursive formula for Recursive Least Squares (RLS) is as follows:

[0143]

[0144]

[0145] Where K(k4) is the recursive least squares (RLS) gain vector, P(k4) is the covariance matrix, P(k4-1) is the covariance matrix of the previous step, and φ(k4) is the regression vector (composed of the current state and the input). The vector is the transpose of the regression vector, λ = 0.995 is the forgetting factor, θ(k4) is the parameter estimate for the current step, θ(k4-1) is the parameter estimate for the previous step, Δy(k4) is the bias signal, and k4 is the discrete-time index (time step number) used to represent the value of the variable at different sampling times. The single-step correction step size is limited to ±0.01 to prevent abrupt parameter changes. The corrected parameters are used to update the discrete system matrix. The corresponding elements in the matrix. Each recursive least squares (RLS) correction performs 10 recursive iterations, with a total time of approximately 200 μs.

[0146] Regarding the coupling protection between calibration and rainflow counting, the recursive least squares (RLS) parameter update can cause a step jump in the temperature sequence output by the digital twin. If this jump is captured by the rainflow counting submodule, it will generate a false large-value thermal cycle count, leading to an artificial increase in the cumulative damage. Therefore, this embodiment designs the following coupling protection mechanism: when the recursive least squares (RLS) parameter update is triggered, module 5 sets the internal flag CORRECTION_ACTIVE=1 and records the calibration time. The rainflow counting submodule checks this flag before each execution. If CORRECTION_ACTIVE=1, the rainflow counting submodule will compare the current time with... Comparison: If the current time is in [ 5s Within a transition window of +5s, counting is paused and the temperature data within that window is marked as invalid. After the current time exceeds the transition window, the CORRECTION_ACTIVE flag is cleared and normal counting resumes. The transition window length is set to 10 seconds because: the step size limit of the recursive least squares (RLS) correction is ±0.01, corresponding to a maximum change in thermal resistance of approximately 0.01°C / W, a temperature jump of approximately 1°C to 2°C at typical power levels (100W to 200W), and a thermal time constant of approximately 3 to 5 seconds for the digital twin model. Therefore, a 10-second window is sufficient to cover the process of the model output reconverging to a new steady state after parameter updates.

[0147] In the rainflow counting, the rainflow counting submodule performs the standard four-point rainflow counting method on the digital twin junction temperature time series.

[0148] The theoretical lifetime of each thermal cycle is calculated using a modified Coffin-Manson model, which considers both the thermal cycle amplitude and mean temperature. The theoretical lifetime of each thermal cycle is calculated using a modified Coffin-Manson-Arrhenius model.

[0149] in, This is the theoretical fatigue life corresponding to the i-th thermal cycle, A1 = 3.5 × 10¹4 For lifespan factor, It is the junction temperature fluctuation amplitude of the i-th thermal cycle, and r1=4.5 is the Coffin-Manson exponent. =0.8eV is the activation energy (a typical value for fatigue of the solder layer in SiC power modules). =8.617×10 -5 eV / K is the Boltzmann constant. Here, Kelvin represents the cycle-average junction temperature. The introduction of the Arrhenius correction term results in a shorter theoretical lifetime for thermal cycles operating at high average temperatures, which aligns with the physical reality of accelerated device degradation under the high base temperature conditions of the Shago Desert summer.

[0150] Based on this, the cumulative damage is calculated according to the linear accumulation rule:

[0151] Where D represents the total cumulative damage, and i represents the i-th thermal cycle condition. Let i be the actual number of occurrences of the i-th type of thermal cycle. This corresponds to the theoretical lifespan. When the junction-to-shell equivalent thermal resistance... When the cumulative damage increases by more than 20% compared to the initial value (indicating significant degradation of the solder layer), the cumulative damage is multiplied by an acceleration factor of 1.5 to reflect the accelerated degradation effect. A warning signal is sent when D>0.7, and the FSM switches to S4 derating mode when D>0.9.

[0152] The multi-failure mode damage fusion includes a weighted fusion of solder layer fatigue damage, bond wire lift-off damage, and package delamination damage.

[0153] The fatigue damage degree of the solder layer, also known as the total cumulative damage degree D, reflects the degree of crack propagation in the solder layer under thermal cycling stress. Its physically observable degradation symptom is the junction-to-shell thermal resistance. The increase. This embodiment uses online monitoring. The trend was used to cross-validate the fatigue damage of the solder layer. The accuracy.

[0154] The bond line separation damage degree Estimation is based on on-resistance degradation. In SiC MOSFET packages, aluminum bond wires experience foot crack propagation under thermal cycling stress, ultimately leading to increased resistance and lift-off failure. This embodiment utilizes online ESR (Equivalent Series Resistance) estimation to obtain the on-resistance trend of the channel and calculates the bond wire degradation index:

[0155] in It is the on-resistance deviation at time k5. It is the measured on-resistance at time k5. The theoretical on-resistance is based on the current junction temperature (calculated using a temperature coefficient model). This is the estimated virtual junction temperature at time k5. When the bond wire degrades, the actual on-resistance will systematically exceed the temperature model prediction. Bond wire damage is defined as:

[0156] in It is the degree of bond line separation damage. The on-resistance failure threshold is defined in the device datasheet (typically a 20% increment of the initial value, i.e., 13mΩ in this embodiment). The system... A 24-hour moving average was used to eliminate the effects of thermal cycling and measurement noise.

[0157] The aforementioned encapsulation layer damage degree The estimation is based on an abnormally increased thermal resistance from the heatsink to the environment. Degradation of the solder layer between the DBC (Direct Bonded Copper) and the copper base plate, or detachment of the adhesive layer, prevents heat from being effectively conducted to the heatsink, resulting in an increased equivalent thermal resistance from the heatsink to the environment. The damage increases and cannot be recovered by cleaning the heatsink. The degree of package delamination damage is defined as:

[0158] in It is the degree of damage caused by encapsulation layering. This is the currently estimated thermal resistance from the heatsink to the environment. The dust degradation coefficient (calculated by the FSM module, reflecting the increase in thermal resistance caused by recoverable external contamination). The thermal resistance from the radiator to the environment is the factory-specified value. The failure thermal resistance threshold for the package layer (set to 2.5 times the factory calibration value). By removing the dust accumulation degradation component, this indicator only reflects irreversible internal package degradation.

[0159] The weighted fusion method calculates the comprehensive health index HI by weighting the independent damage levels of the three failure modes.

[0160] Among them, weight = 0.5, = 0.3, = 0.2, reflecting engineering experience that solder layer fatigue is the dominant failure mechanism of SiC (Silicon Carbide) power modules, bond line degradation is a secondary mechanism, and package layering is an auxiliary mechanism. The sum of the weights is 1. The value range of HI is [0, 1], where 0 represents complete health and 1 represents reaching the comprehensive failure threshold.

[0161] Furthermore, the comprehensive health index also applies a short-board effect correction: if any single damage level exceeds 0.9 (i.e., any failure mode is close to the critical state), then regardless of the weighted fusion result, HI is directly set to max(HI, 0.9). This ensures that the critical degradation of any single failure mode can trigger a system-level protection response.

[0162] The aforementioned prediction of health indicator degradation trends includes degradation trend fitting. In terms of degradation trend fitting, the DTHM module maintains a historical sequence buffer for health indicators, storing the average daily HI values ​​for the most recent 30 days (30 data points in total). Weighted least squares method is used to perform linear trend fitting on the HI time series.

[0163] in These are the predicted health indicators at time t. For the fitting intercept, The slope of the degradation trend (in units of 1 / day) is given, and t is the time in days. In the weighted least squares, the weight of recent data (last 7 days) is twice that of older data (more than 7 days ago) to enhance the sensitivity of the trend fit to the latest degradation dynamics.

[0164] The remaining useful life prediction includes a nominal remaining useful life (RUL) prediction, which is calculated by extrapolation based on the slope of the degradation trend.

[0165] in =1.0 is the health indicator failure threshold, HI( ( ) represents the current health indicator value. The slope of the degradation trend. When When ≤0 (health indicators do not show a deterioration trend), Set as the upper limit of the system's design life.

[0166] The uncertainty of the RUL prediction is quantified by the moving standard deviation σλ of the degradation slope. The system also calculates the optimistic remaining useful life. and pessimistic remaining lifespan : (Taking the case of slowed degradation) (Taking the case of accelerated degradation) All three RUL values ​​(nominal, optimistic, and pessimistic) are written into the health status subvector and reported to the remote operation and maintenance platform through the communication interface, providing a quantitative basis for maintenance decisions.

[0167] In the predictive maintenance decision support, the DTHM module outputs the following maintenance decision suggestions to the FSM module and the remote operation and maintenance platform based on comprehensive health indicators: Level 1 warning (HI>0.5 or <180 days): The system will report a recommendation via the communication interface to schedule a preventative check during the next planned shutdown. The FSM module will not impose any operational restrictions.

[0168] Level II warning (HI>0.7 or <90 days or any single damage level>0.7): If the communication interface is used to report and it is recommended to arrange maintenance within 30 days, the FSM module will lower the derating initiation temperature by 5°C as a preventive derating.

[0169] Level 3 warning (HI>0.9 or <30 days or any single damage level>0.9): Report an emergency maintenance request through the communication interface, and the FSM module switches to S4 derating mode to limit the maximum power to 70% of the rated value.

[0170] Failure alarm (HI≥1.0): The FSM module is migrated to S6 and is immediately disconnected and locked. It can only be restarted after manual inspection and confirmation.

[0171] The impedance environment sensing module performs a complete impedance spectrum scan and updates the environment state subvector when the system is in standby or the load rate is below 10%. When the system is in normal operation and the load rate is consistently above 10%, the narrowband fast impedance detection mode is enabled.

[0172] Module 6 (Impedance Environment Sensing Module IBEP) performs a full impedance spectrum scan and updates the environmental state sub-vector when the system is in standby or the load rate is below 10%. Furthermore, when the system is in normal S3 operation and the load rate remains above 10%, Module 6 activates a narrowband fast impedance detection mode, injecting a small common-mode disturbance signal at a single 100kHz frequency point using the load transient gap, acquiring the response, and calculating the impedance magnitude at that frequency point. If this magnitude deviates by more than 20% compared to the previous full scan, Module 6 sets an alarm flag and increases the priority of the full scan.

[0173] In impedance spectrum measurement, when the system is in standby or the load rate is below 10%, module 6 performs a complete impedance spectrum scan according to the following steps: First, a common-mode disturbance signal is superimposed on the symmetrical PWM (Pulse Width Modulation) signal using a full-bridge or half-bridge drive circuit. This disturbance signal is a logarithmically swept sine wave from 10kHz to 1MHz, with an amplitude of 1% to 3% of the bus voltage (typically 10V), and a sweep time of 5 seconds. Second, the common-mode response current is acquired through a high-frequency common-mode current transformer (bandwidth DC to 10MHz, sensitivity 100mV / mA) connected in series at the common terminal, using an independent 12-bit 5MSPS high-speed ADC (analog-to-digital converter). Third, a 512-point DFT (Discrete Fourier Transform) is performed on the excitation voltage and response current to calculate the complex impedance at each frequency point.

[0174] For narrowband fast impedance detection, when the system is in S3 operation and the load rate is consistently above 10%, module 6 activates the narrowband fast impedance detection mode. This mode utilizes load transient gaps (such as the 50ms recovery window after an MPPT (Maximum Power Point Tracking) power step) for fast detection at a single frequency point. Module 6 continuously monitors the rate of change of the load current. When it detects that the rate of change has fallen from a high value to a steady-state level, it waits 20ms to allow the output filter to fully establish a new steady state. Then, within the next 30ms, it injects a small common-mode disturbance with an amplitude of 0.5% of the bus voltage into the 100kHz frequency point and acquires the response. The impedance magnitude at this frequency point is calculated and compared with the corresponding value from the previous complete scan. If the deviation exceeds 20%, an alarm flag is set.

[0175] Regarding environmental condition identification and the linkage control of self-cleaning and dehumidification, the environmental condition identification criteria are defined as follows: Clean state: The impedance modulus in the low-frequency band is greater than 80% of the calibrated value, and the phase angle across the entire frequency band is within... 80° to 90°. Dry and dusty condition: The imaginary part of the impedance in the high-frequency band deviates by more than 15% of the rated value, while the impedance modulus in the low-frequency band does not change significantly. Moist and dirty condition: The impedance modulus in the low-frequency band is lower than 50% of the rated value, and the phase angle deviates towards 0°.

[0176] When a dust accumulation alarm is triggered, module 6 sends a cleaning command to the piezoelectric ceramic drive circuit. After cleaning, a second scan is performed to verify: if the impedance recovers to more than 90% of the cleaning standard, it is considered effective; otherwise, the process is repeated once. If both scans are ineffective, a maintenance alarm is sent.

[0177] When a moisture and contamination alarm is triggered, the system must not start high-pressure operation and will automatically enter the S1 drying and dehumidification mode. ATSM insulation (low duty cycle inductor circulating current low-power controlled heating) is used to raise the PCB board temperature to 40°C to 60°C to evaporate moisture. During dehumidification, impedance spectrum measurement is performed every 5 minutes to monitor recovery. When the low-frequency impedance modulus recovers to above 70% of the calibrated value and the phase angle recovers to [value missing], [the system will proceed]. When the temperature is below 60°C, dehumidification is considered complete.

[0178] The intelligent protection state machine module manages the system's operating status and makes decisions on protection actions based on the following information: temperature information in the temperature state sub-vector; cumulative damage degree, thermal resistance information, and comprehensive health indicators in the health state sub-vector; environmental state in the environmental state sub-vector; and current control parameters in the control state sub-vector.

[0179] Module 7 (Intelligent Protection State Machine Module FSM) reads temperature information from the temperature state subvector, cumulative damage level, thermal resistance information, and comprehensive health indicators from the health state subvector, environmental state from the environmental state subvector, and current control parameters from the control state subvector. It then comprehensively judges and executes state transitions. When the FSM module undergoes a state transition, it sets a flag in the mode subvector. All other modules check this flag and perform mode reconfiguration at the beginning of each intermediate frequency cycle.

[0180] The decision-making for the management and protection actions of the system's operating status includes: judging and executing operating status transitions; adaptive dynamic derating for dust accumulation; and recovery from abnormal operating status.

[0181] The FSM module defines seven operating states: S0 Initialization state (power-on self-test and initial environmental assessment), S1 Preheating state (ATSM module insulation or IBEP module dehumidification), S2 Soft start state (current limiting and voltage ramping), S3 Normal operation state (full-power FNN-PID adaptive control), S4 Dynamic derating state (power limiting triggered by temperature or lifespan), S5 Transient ride-through state (short-term overload allowed), and S6 Emergency cut-off state (hardware failure or absolute temperature exceeding limit).

[0182] Regarding the judgment and execution of operating state transitions, during the transition from the S3 normal operating state to the S5 transient ride-through state, when the FSM module detects that the load current has entered the 120% to 200% rated value range, it initiates transient duration prediction logic. This logic classifies and judges based on the load current change rate: if its absolute value is greater than 10% per millisecond of the rated current (i.e., the load current is changing rapidly), it is judged as a transient event, allowing the transition to the S5 transient ride-through state and starting a 10-second countdown window; if its absolute value is less than 2% per millisecond of the rated current and lasts for more than 500ms (i.e., the load current tends to stabilize in the overload range), it is judged as a continuous overload rather than a transient event, and the FSM module skips the S5 transient ride-through state and directly triggers the transition to the S4 dynamic derating state. Upon entering the S5 transient ride-through state, the FSM module continuously monitors the load current: if the current drops below 120% of the rated value within a 10-second window, it returns to the S3 normal operating state; if the current is still above 120% of the rated value when the 10-second window expires, the FSM module immediately transitions from the S5 transient ride-through state to the S4 dynamic derating state. This two-level decision mechanism ensures that the S5 transient ride-through state is only used for true short-term transient overload scenarios, avoiding misjudging continuous overloads as transients and delaying protection response.

[0183] When the FSM module undergoes a state transition, it executes the following broadcast and handshake process: The FSM module updates the current state in the system mode subvector to the new state value, simultaneously updates the configuration, and clears the ACK (acknowledgment) bits of all modules in the configuration acknowledgment register. Upon detecting the configuration update at the start of the next intermediate frequency cycle, each functional module performs mode reconfiguration: Module 2 loads the basic PID (Proportional-Integral-Derivative) parameter set for the corresponding state (in S2 soft-start state). The benchmark value of the scaling factor is reduced by 40%. The baseline value of the integral coefficient is reduced by 60%; under S4 dynamic derating status. The baseline value of the proportional gain is reduced by 30% to avoid oscillations during derating; S5 transient crossover state The proportional gain is increased by 20% to enhance overload transient response; module 3 switches the dead time upper and lower limits according to the new state; module 4 assesses whether the ATSM module's operating mode needs to be changed; module 5 records the operating condition classification label in the health log and updates the degradation trend weighting. After each module completes reconfiguration, it writes its own ACK (acknowledgment) bit. After the FSM module detects all ACK bits set to 1 in the next intermediate frequency cycle, it clears the configuration update flag. If the handshake times out (if no module responds after 10 consecutive intermediate frequency cycles), the FSM module records a configuration anomaly alarm but does not block operation.

[0184] Regarding adaptive dynamic derating due to dust accumulation, the FSM module continuously calculates the equivalent heatsink-to-ambient thermal resistance and maintains a 24-hour moving average. When the dust degradation coefficient meets a set threshold, the derating initiation temperature decreases from the default value, triggering a piezoelectric self-cleaning and maintenance alarm.

[0185] Regarding asymmetric hysteresis recovery, recovery from S4 dynamic derating state or S6 emergency shutdown state requires not only that the temperature drop below the shutdown threshold, but also that the cooling rate condition be met and maintained for a certain period of time. An additional forced cooling waiting time is added when recovering from S6 emergency shutdown state.

[0186] Regarding recovery from abnormal operating conditions, the recovery path from the S6 emergency shutdown state involves forced cooling, recovery condition assessment, recovery self-test, DTHM module damage reassessment, and safe soft start, before directly transitioning to the S1 preheating state (if dehumidification is required) or the S2 soft start state (if conditions are met, direct soft start), and then proceeding through the normal process to enter the S3 normal operating state. Phase 1, forced cooling period: all PWM (Pulse Width Modulation) outputs remain blocked, the real-time core only runs temperature acquisition and observer updates, does not perform control law calculations, and the DTHM module continues to run and record the cooling curve.

[0187] Phase Two, Recovery Condition Assessment; After the forced cooling period ends, the FSM module checks: whether the junction temperature has dropped below 135°C (i.e., the S6 emergency shutdown trigger temperature of 150°C minus the hysteresis of 15°C); whether the temperature change rate is continuously negative and the absolute value is greater than 0.5°C / min for 30 seconds; both conditions must be met simultaneously to proceed to the next phase.

[0188] Phase 3: The IBEP module resumes self-test; the FSM module triggers module 6 to perform a complete impedance spectrum scan to eliminate possible PCB (Printed Circuit Board) insulation damage caused by the S6 emergency cut-off state fault. If moisture accumulation is detected, it enters the S1 preheating state to perform dehumidification and then resumes. If the dust accumulation is detected as clean or dry, it continues to the next phase.

[0189] Phase 4, DTHM module reassessment of damage; FSM module reads the cumulative damage and comprehensive health index from the health status sub-vector. If the cumulative damage... If the threshold is >0.9 or the comprehensive health index HI>0.9, the system will revert to the S4 dynamic derated state instead of the S3 normal operation state at full power. If neither of these thresholds is exceeded, the system will enter the S2 soft start state.

[0190] Phase 5, Safe Soft Start: Follow the standard soft start procedure. If the S6 emergency cut-off state is triggered by overcurrent, the soft start current limit is further reduced to 30% of the rated value. If the S6 emergency cut-off state is triggered by temperature exceeding the limit, the soft start adopts the conventional 50% current limit, but the voltage ramp time is extended to 3 times that at room temperature.

[0191] The line voltage drop compensation module calculates the voltage drop compensation amount of the remote cable in real time based on the temperature information in the temperature state sub-vector and the load current obtained by the power level sampling circuit.

[0192] Module 8 (LDC) reads the ambient temperature from the temperature state subvector, obtains the load current from the power stage sampling circuit, and calculates the cable voltage drop compensation amount according to the compensation formula. After being clamped at the upper limit, the compensation amount is superimposed on the voltage control reference of Module 2. Module 8 updates the compensation amount and writes it to a dedicated field of the control state subvector in each intermediate frequency adaptive adjustment cycle. Module 2 reads the adjusted reference voltage to replace the fixed nominal value when calculating the PID (Proportional-Integral-Derivative) error in the high frequency control cycle.

[0193] The formula for calculating the compensation voltage in line voltage drop compensation is:

[0194] in It is a compensation voltage. To measure the load current, The resistance of the cable is 0.00393 at 25°C, where 0.00393 is the temperature coefficient of resistance. 25 is the current ambient temperature, and 25 is the reference temperature. The LDC module calculates the compensation amount and writes it into the control state sub-vector in each intermediate frequency control cycle. This feedforward compensation does not change the characteristic equation of the PID (Proportional-Integral-Derivative) closed-loop transfer function.

[0195] This embodiment also proposes a control method for an adaptive wide-temperature-range power supply controller, such as... Figure 2 As shown, the method is used to control an adaptive wide-temperature-range power controller system. The method includes system initialization, high-frequency real-time control, medium-frequency adaptive adjustment, and low-frequency health management and environmental perception: System initialization: The system hardware performs a self-test and initializes parameters and operating modes; High-frequency real-time control: When the intelligent protection state machine module enters the soft-start state or normal operation state, the controller system cyclically executes high-frequency control; Medium-frequency adaptive adjustment: The controller system cyclically executes medium-frequency adaptive adjustment; Low-frequency health management and environmental perception: The controller system executes low-frequency health management and environmental perception in a non-preemptive manner.

[0196] The controller system cyclically performs high-frequency control, including cyclically performing high-frequency control at a rate of 100 kHz (period 10 μs). The controller system cyclically performs intermediate-frequency adaptive adjustment, including cyclically performing intermediate-frequency adaptive adjustment at a rate of 1 kHz (period 1 ms). The controller system performs low-frequency health management and environmental perception in a non-preemptive manner, including performing low-frequency health management and environmental perception in a non-preemptive manner at frequencies from 1 Hz to 0.1 Hz.

[0197] The system initialization phase includes: performing a self-test on the system hardware; acquiring physical temperature through a thermistor sensor; rapid impedance self-testing by the impedance environment sensing module to quickly determine the environment category; digital twin initialization of the edge digital twin health management module to verify whether the deviation between the initial output of the digital twin and the physical measurement is within a reasonable threshold range; and making an initial working mode decision to determine the working mode of the module.

[0198] The initial operating mode decision, used to determine the module's operating mode, includes: if the ambient temperature is lower than the set temperature value or the environment is damp and contaminated, the intelligent protection state machine module enters a preheating state; if the ambient temperature is not lower than the set temperature value and the environment is not damp or contaminated, the intelligent protection state machine module enters a soft-start state; if the comprehensive health index loaded by the edge digital twin health management module is greater than the set value, then regardless of temperature and environmental conditions, the intelligent protection state machine module enters a dynamic derating state to protect the severely degraded power module. The set temperature value is... 25°C; the set value is 0.9.

[0199] The high-frequency real-time control includes the controller system cyclically executing high-frequency control after the intelligent protection state machine module enters the soft-start state or normal operation state. This includes the controller system cyclically executing the following high-frequency control at a rate of 100kHz (period 10μs): synchronous data sampling: acquiring data such as voltage, current, and on-state voltage drop of the controller; virtual junction temperature estimation update: calculating the junction temperature by substituting the sampled on-state voltage drop value into the calibration equation, and updating the virtual junction temperature estimate after 8-point moving average filtering; online ESR (Equivalent Series Resistance) estimation update: calculating and updating the online ESR (Equivalent Series Resistance) estimate based on the voltage AC ripple component and the inductor current ripple component; multi-source information fusion observer iteration: performing one-step iteration of the discrete state equation of the multi-source information fusion observer; PID (Proportional-Integral-Derivative) control law calculation: including voltage error calculation, where the reference voltage is the value compensated by the line voltage drop compensation module; hardware protection monitoring: if the inductor current or output voltage exceeds the rated threshold range, the hardware comparator immediately triggers an interrupt and forces the intelligent protection state machine module to the emergency disconnect state.

[0200] The intermediate frequency adaptive regulation includes the controller system cyclically executing the following intermediate frequency adaptive regulation: fuzzy inference based on temperature data to weighted synthesize PID adjustment coefficients; synthesizing the final PID (Proportional-Integral-Derivative) parameters through a neural network, while simultaneously performing backpropagation to update network weights; mode arbitration and execution of the active thermal stress management module: the active thermal stress management module reads the request flag and the state of the intelligent protection state machine module, determines whether to accept the request according to the arbitration priority, and executes the corresponding mode switch; PID (Proportional-Integral-Derivative) parameter linkage adjustment during frequency modulation: if currently in the frequency modulation mode of the active thermal stress management module, the PID (Proportional-Integral-Derivative) parameter is adjusted accordingly. The phase margin is scaled twice to maintain a phase margin of 45° or higher. Dead time update: The estimated virtual junction temperature is read from the temperature state sub-vector, and the temperature-compensated dead time is calculated and updated. Digital twin iteration: A one-step iteration of the reduced-order state equation is performed. Compensation calculation for the line voltage drop compensation module: The voltage drop compensation amount of the remote cable is calculated, and the reference voltage is updated and adjusted after upper limit clamping for high-frequency cycle use. Intelligent protection state machine module state transition judgment: The comprehensive health index is checked to see if it exceeds the warning threshold, which refers to the level 3 warning threshold of 0.9. If so, a transition to dynamic derating state is triggered. Double-buffered pointer toggling: The read / write roles of two equal-length storage areas are toggled through atomic operations.

[0201] The low-frequency health management and environmental perception include the controller system performing the following low-frequency health management and environmental perception in a non-preemptive manner: performing thermistor sensor sampling and thermal field reconstruction at a frequency of 1Hz: performing filtered sampling on the thermistor sensor and substituting it into a cubic spline interpolation algorithm to calculate an 8×6 grid temperature distribution map; performing thermal resistance trend analysis at a frequency of 0.5Hz: calculating the dust accumulation degradation coefficient. If the change in the dust accumulation degradation coefficient exceeds the set dust accumulation degradation coefficient (which is 0.05), the derating initiation temperature is dynamically adjusted and the intelligent protection state machine module is notified. Rainflow counting and damage update are performed at a frequency of 0.1Hz: Rainflow counting is performed on the junction temperature sequence of the most recent 10 seconds, and the theoretical lifetime is calculated and the cumulative damage is updated after extracting the thermal cycle. Here, the theoretical lifetime is calculated and the cumulative damage is updated according to the modified Coffin-Manson-Arrhenius model. Multi-failure mode damage fusion is performed at a frequency of 0.1Hz: After completing the damage update in the previous step, the bond wire lift-off damage and package delamination damage are calculated, and then the bond wire lift-off damage, package delamination damage, and solder layer fatigue damage are weighted and fused to obtain a comprehensive health index. Here, the weights are (0.5, 0.5, 0.05). 3.0.2) Weighted fusion of bond line separation damage, encapsulation layer damage, and solder layer fatigue damage to obtain a comprehensive health index; Hourly degradation trend fitting and remaining useful life (RUL) prediction: Read the historical sequence of the daily average value of the comprehensive health index for the most recent 30 days, use weighted least squares to fit a linear degradation trend, and extrapolate the nominal, optimistic, and pessimistic values ​​of remaining useful life based on the degradation slope; Impedance spectrum measurement of the impedance environment sensing module; Online model correction of the edge digital twin health management module: Compare the digital twin output with the observer output every hour or after a change in operating conditions. When the deviation exceeds the set deviation value for 10 consecutive times (the set deviation value is 3°C), the recursive least squares incremental correction is initiated; Hourly health log persistence: Write a snapshot of the complete health status at the current moment to non-volatile storage.

[0202] Furthermore, the method also includes state transition linkage control, which is an event-driven process triggered by the intelligent protection state machine module within the intermediate frequency adaptive adjustment cycle. The state transition linkage control includes a linkage process for the intelligent protection state machine module to transition from a normal operating state to a dynamic derating state, and a linkage process for recovery from an emergency disconnection state.

[0203] The linkage process for transitioning from normal operation to dynamic derating includes: The FSM module detects that the junction temperature exceeds the derating initiation temperature, or the comprehensive health index HI exceeds 0.9. The FSM module sets its state to S4 dynamic derating state. Each module detects and performs reconfiguration in the next intermediate frequency cycle: Module 2 loads the S4 dynamic derating state basic parameter set; Module 4 evaluates whether to simultaneously initiate frequency modulation; Module 5 marks the health log condition as derating operation and records the trigger reason. The FSM module calculates the current maximum power according to the derating curve and implements power limiting.

[0204] The recovery process from the emergency shutdown state includes: after the S6 emergency shutdown state is triggered, all PWM (Pulse Width Modulation) modules are locked, module 5 records the fault, and a forced cooling period begins. After the cooling period, the FSM module checks the temperature conditions. The FSM module triggers module 6 to perform a complete IBEP module impedance spectrum scan. If moisture or dirt accumulation is detected, it enters the S1 preheating state for dehumidification; if cleanliness or dryness of the accumulated dust is detected, it continues. The FSM module reads the cumulative damage degree and comprehensive health index from the health status sub-vector, and determines whether to enter the S4 dynamic derating state or the S2 soft start state based on the cumulative damage degree or comprehensive health index. A safe soft start is performed, and after successful start, the FSM module migrates to the S3 normal operation state.

[0205] The aforementioned control methods form a nested loop structure on the time scale. High-frequency real-time control ensures basic voltage regulation and transient response capability of the output voltage. Mid-frequency adaptive regulation reads updated temperature information from the high-frequency layer during each execution and outputs adaptive PID (Proportional-Integral-Derivative) parameters for use by the high-frequency layer, simultaneously driving the decision logic of the ATSM and FSM modules and the real-time simulation of the digital twin model. Low-frequency health management acquires the complete state of the mid-frequency layer through a double-buffering mechanism during each execution, performing trend analysis, multi-mode damage fusion, degradation prediction, and predictive maintenance calculations. State transition linkage control is an event-driven process triggered by the FSM module within the mid-frequency cycle.

[0206] Deterministic data transfer is achieved between the components through a global state vector and a double-buffering mechanism. The DTHM module spans both mid-frequency and low-frequency time layers: real-time derivation of the reduced-order model is performed at the mid-frequency layer to ensure synchronous tracking with the physical system, while background tasks such as parameter correction, rainflow counting, damage fusion, trend prediction, and log persistence are performed at the low-frequency layer to fully utilize the application core's computing resources. This layered architecture ensures the executability of the control method on computationally limited embedded MCUs (Microcontroller Units), while also guaranteeing data consistency and timing determinism between layers, enabling those skilled in the art to fully implement the control method of this invention.

[0207] This patent also proposes embodiments in the following scenarios: 1. In low-temperature winter environments: Ambient temperature at 20:00 The temperature dropped from 5°C and continued to decrease. The ATSM module detected that the junction temperature was below the insulation start threshold (0°C) and initiated insulation mode. Module 5's digital twin simulation engine switched to insulation power calculation mode. The effective value of the inductor circulating current was approximately 12.5A, corresponding to a heating power of approximately 25W, maintaining the junction temperature at 20°C. After sunrise at 06:30, the ambient temperature rose to 10°C, the ATSM exited insulation mode, and module 5 resumed standard power calculation. The FSM module entered S2 soft-start state, with a current limit of 17.5A and a voltage ramp of 200ms. At 07:05, the output stabilized, the FSM module migrated to S3 normal operation state, and each module loaded the S3 parameter set. Module 5 recorded a health log snapshot.

[0208] 2. Full load during high summer temperatures: At 12:00, the ambient temperature was 45°C, the rack temperature was 80°C, the load rate was 90%, and the junction temperature was 108°C. The ATSM module initiated frequency modulation, adjusting the frequency to 70kHz. Module 2 synchronously scaled the PID (Proportional-Integral-Derivative) parameters. Module 5 switched power calculation to frequency modulation mode, continuously comparing the digital twin output with the physical measurement values. The junction temperature dropped to 95°C, and the system operated stably at 85% power. At 14:00, the load dropped to 30%, the junction temperature dropped below 65°C, the ATSM module exited frequency modulation, and all parameters returned to normal. Module 5 logged the data; the cumulative damage increment for the day was 0.0003.

[0209] 3. Condensation and recovery after a sandstorm: At 07:00 the day after the sandstorm, a full scan of the IBEP module detected that the low-frequency impedance modulus had dropped to 40% of the calibrated value, and the phase angle had deviated to... 25°C, indicating moisture and dirt accumulation. The FSM module prevents high-voltage startup and enters the S1 preheating dehumidification mode. The ATSM module heats the PCB (Printed Circuit Board) to 50°C using a low duty cycle inductor circulating current (approximately 13.7A RMS inductor current, approximately 30W heating power). The IBEP module scans and monitors every 5 minutes. Module 5 continuously runs digital twin and records thermal cycling data during dehumidification. At 09:30, the impedance recovers to over 75% of the calibrated value, and the phase angle recovers to... 65°C, dehumidification complete. The FSM module enters normal operation mode (S3) after a soft start (S2), and the entire module is reconfigured. Module 5 records dehumidification events and their durations to the health log.

[0210] 4. Overcurrent fault triggers S6 emergency shutdown state and complete recovery: When the system is in normal S3 operation, a short circuit fault in a downstream device causes the load current to surge to 250% of the rated value. The hardware overcurrent comparator triggers an interrupt within 0.5μs, blocking all PWM (Pulse Width Modulation) outputs. The FSM module immediately transitions to the S6 emergency shutdown state. Module 5 records a fault snapshot (including the current comprehensive health index HI=0.45, various damage levels, and estimated RUL values) and continuously records the cooling curve.

[0211] Phase 1: A 60-second forced cooling period during which the junction temperature continuously decreases from 112°C. The digital twin simulation engine in Module 5 continues to run, verifying whether the cooling dynamics match the model's expectations.

[0212] Phase Two: After the cooling period, the junction temperature of the FSM module is measured to be 68°C < 135°C, and the temperature change rate is... The temperature condition is met if the temperature is maintained at 1.2°C / min for more than 30 seconds.

[0213] Phase 3: The FSM module triggers module 6 to perform a complete IBEP impedance spectrum scan. The scan results show a slight decrease in the impedance modulus in the low-frequency range (down to 85% of the calibration value), but the phase angle remains unchanged. 75°, determined to be dry dust accumulation, not insulation damage. Scan passed.

[0214] Phase 4: The FSM module reads the cumulative damage from the health status sub-vector. =0.45<0.9, the comprehensive health index HI=0.45<0.9, no reduction protection is needed.

[0215] Phase 5: The FSM module enters S2 soft-start state. Because the S6 emergency shutdown state is triggered by overcurrent, the current limit is set to 30% of the rated value (10.5A), and the voltage ramp time is set to 300ms. The output stabilizes at 48V, and the FSM module transitions to S3 normal operation state. Module 5 records the complete recovery event to the health log. The entire process, from fault to recovery to normal operation, takes approximately 4 minutes.

[0216] In summary, the key points of the technical solution of this invention patent include: 1. Edge Digital Twin Health Management (DTHM) and Coupled Protection Mechanism of Parameter Correction and Rainflow Counting: The primary key protection point of this invention lies in proposing a digital twin health management architecture that operates in real-time at the microcontroller edge, along with its unique anti-spurious damage counting mechanism. The system utilizes an intrinsic orthogonal decomposition (POD) reduced-order thermal model for real-time temperature extrapolation and corrects thermal resistance parameters online using recursive least squares (RLS). Its most significant innovation is the pioneering coupling protection mechanism between correction and rainflow counting to prevent abrupt temperature sequence jumps caused by RLS parameter updates from being misinterpreted as large-scale thermal cycles by the rainflow counting module. When parameter correction is triggered, the system automatically sets an internal correction flag and pauses rainflow counting within a 5-second transition window before and after the correction, marking the data within this window as invalid, thus completely avoiding the spurious increase in accumulated damage. Furthermore, this module independently quantifies three independent failure modes—solder layer fatigue, bond line lift-off, and package delamination—and weights and fuses them into a comprehensive health index (HI). It also introduces a short-board effect correction (i.e., if any single damage level exceeds the limit, the overall index is set to critical), which is also an important protection target for subsequent predictive maintenance decisions.

[0217] 2. Active Thermal Stress Management (ATSM) and its arbitration linkage mechanism with adaptive control: This invention overcomes the limitations of traditional power supplies that rely solely on passive cooling. Its second key protection point is the design of an active thermal management strategy that includes nighttime heat preservation and high-temperature frequency modulation peak clipping, along with supporting multi-level arbitration logic. Its unique technical feature lies in the innovative use of low-duty-cycle inductor circulating current heating instead of external heaters for nighttime heat preservation. In heat preservation mode, the system disconnects the load and operates at an extremely low duty cycle of 0.5% to 5% using the main power Buck topology. This converts the I²R losses generated by the inductor current across the MOSFET on-resistance and the inductor's DC resistance into heat energy, achieving controllable and uniform heating of the power components. Simultaneously, this active thermal management is triggered by heat preservation or peak clipping request flags generated by the fuzzy rule follower of the fuzzy neural network PID (FNN-PID) control module. The ATSM module not only receives these requests but must also perform strict priority arbitration based on the current operating mode of the intelligent state machine (e.g., ignoring certain requests in emergency shutdown or derating states), thereby ensuring that the thermal management actions do not conflict with system-level safety protection.

[0218] 3. Hardware linkage mechanism between impedance environmental sensing (IBEP) and dehumidification self-cleaning: The third key aspect of this invention focuses on online sensing and self-recovery linkage technology for harsh physical environments. This solution reuses the main power supply topology, achieving full-band impedance spectrum scanning or narrow-band fast impedance detection by superimposing a logarithmic sweep sine wave or a single-frequency micro common-mode disturbance onto a symmetrical PWM signal. Its core protection feature lies in the accurate identification of environmental conditions and the forced linkage strategy: the system can classify the environment into three states—"clean," "dry with dust," and "moist with dirt"—based on the extracted impedance modulus and phase angle characteristics. More importantly, when "moist with dirt" is detected, the system will forcibly lock the high-voltage startup process through an intelligent protection state machine, automatically switch to a drying and dehumidification mode, and link the aforementioned low duty cycle inductor circulating current heating mechanism to raise the PCB temperature to evaporate moisture. If "dry with dust" is detected, the external piezoelectric ceramic drive circuit will emit ultrasonic waves of a specific frequency for self-cleaning. This hardware-level closed-loop self-recovery mechanism triggered by impedance characteristics is a highly valuable protection point of this patent.

[0219] 4. Multi-stage disaster recovery and adaptive derating logic of the intelligent protection state machine (FSM): The fourth key aspect of this invention lies in redefining the system state transition and protection recovery path under extreme operating conditions. Unlike conventional systems that simply cool down and restart after a fault, this invention designs a rigorous five-stage asymmetric hysteresis recovery path for the recovery from the emergency shutdown state (S6) to the normal state. This unique path sequentially includes: a 60-second forced cooling period; a recovery condition assessment with dual thresholds including temperature threshold and sustained negative rate of change; an IBEP impedance self-test to rule out irreversible damage to hardware insulation; a DTHM damage reassessment to confirm whether the device's remaining lifetime requires immediate derating; and finally, a safe soft-start that dynamically adjusts the current limit value based on the triggered fault type. Furthermore, the FSM module also includes a dust accumulation adaptive dynamic derating function, which can dynamically adjust the starting temperature of the triggered power derating based on the online calculated dust accumulation degradation coefficient (i.e., the change ratio of equivalent thermal resistance), effectively avoiding internal thermal runaway of devices under severe dust accumulation.

[0220] 5. Wide-temperature-range multi-source information fusion sensing and cross-core dual-buffer synchronous configuration architecture: The fifth key protection point of this invention covers the underlying hardware and software collaboration and data synchronization architecture supporting all the aforementioned advanced algorithms. To address the data consistency problem of a dual-core microcontroller (real-time core and application core) asynchronously executing multiple tasks at high, medium, and low frequencies, this invention designs a double-buffered interaction mechanism based on a global state vector. Its core technical feature is: two equal-length buffers are allocated in shared RAM. The real-time core writes to the execution read / write buffer at a frequency of 1kHz via a single, extremely short IPC atomic flag, instantly reversing the roles of the two buffers, completely eliminating the risk of the application core reading incomplete "dirty data." Furthermore, the system only truly enters the new state after all distributed modules, including dead time, PID baseline parameters, active thermal management mode, and digital twin power calculation model, have completed parameter reconfiguration for the new state and returned a response, achieving seamless and safe parameter switching under complex control architectures.

[0221] Although specific embodiments of the present invention have been described in detail by way of examples, those skilled in the art should understand that the above examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Those skilled in the art should understand that modifications to the technical solutions described in the above embodiments or equivalent substitutions of some technical features may be made without departing from the scope and spirit of the present invention, and all such modifications or substitutions are within the protection scope of the present invention.

Claims

1. An adaptive wide-temperature-range power supply controller system, characterized in that, The system includes a multi-dimensional temperature sensing module, an adaptive FNN-PID control module, a dead-time dynamic optimization module, an active thermal stress management module, an edge digital twin health management module, an impedance environment sensing module, an intelligent protection state machine module, and a line voltage drop compensation module. The modules interact with each other through the shared RAM area of ​​the dual-core MCU, and a double buffering mechanism is used to ensure cross-core data consistency. The system defines a global state vector as the core data structure shared by all modules. The multi-dimensional temperature sensing module is used to output the system temperature state vector; The adaptive FNN-PID control module is used to output control quantities to the power stage drive circuit; The dead time dynamic optimization module is used to output the optimal dead time to the PWM generator dead time register; The active thermal stress management module is used to execute active thermal control strategies; The edge digital twin health management module is used for predictive maintenance and intelligent health management; The impedance environment sensing module is used to perform impedance spectrum measurement, environmental status identification, and self-cleaning and dehumidification linkage control by utilizing power topology multiplexing. The intelligent protection state machine module is used to integrate information from all modules to manage the system's operating status and make decisions on protection actions; The line voltage drop compensation module is used to calculate the voltage drop compensation amount of the remote cable in real time based on the ambient temperature and load current, and adjust the voltage control reference.

2. The system according to claim 1, characterized in that, The system allocates two equal-length storage areas in the shared RAM for writing and reading data from the global state vector.

3. The system according to claim 1, characterized in that, The global state vector includes a temperature state sub-vector, a control state sub-vector, a health state sub-vector, an environmental state sub-vector, and a system mode sub-vector. The temperature state sub-vector is updated by the multi-dimensional temperature sensing module in each high-frequency real-time control cycle; The control state subvector is updated by the adaptive FNN-PID control module and the dead-time dynamic optimization module in each intermediate frequency adaptive adjustment cycle; The health status sub-vector is updated by the edge digital twin health management module in low-frequency health management and environmental perception. The environmental state sub-vector is updated by the impedance environment sensing module after the impedance spectrum scan is completed; The system mode subvector is updated by the intelligent protection state machine module in each intermediate frequency adaptive adjustment cycle.

4. The system according to claim 3, characterized in that, The multi-dimensional temperature sensing module outputs a temperature state vector to the shared RAM area through physical temperature acquisition, virtual junction temperature estimation, online ESR estimation, and multi-source information fusion.

5. The system according to claim 4, characterized in that, The multidimensional temperature sensing model acquires physical temperature through four thermistor sensors deployed on the controller PCB.

6. The system according to claim 4, characterized in that, The multidimensional temperature sensing model uses the on-state voltage drop signal of the power stage sampling circuit to estimate the virtual junction temperature.

7. The system according to claim 4, characterized in that, The multidimensional temperature sensing model is based on the voltage AC ripple component and inductor current ripple component collected by the system in real time, and obtains the online ESR estimate by calculating the ripple peak ratio.

8. The system according to claim 4, characterized in that, The multi-source information fusion includes at least the fusion calculation of physical temperature acquisition values ​​and virtual junction temperature estimation values.

9. The system according to claim 3, characterized in that, The adaptive FNN-PID control module is responsible for adaptive adjustment of PID parameters based on temperature conditions and calculation of PWM duty cycle, so as to output control quantity to the power stage drive circuit.

10. The system according to claim 9, characterized in that, The adaptive FNN-PID control module performs adaptive adjustment of fuzzy neural network PID parameters and calculation of PWM duty cycle based on the virtual junction temperature estimate and ESR online estimate in the temperature state sub-vector, the current state of the intelligent protection state machine module in the system mode sub-vector, and the working mode of the active thermal stress management module.

11. The system according to claim 3, characterized in that, The dead time dynamic optimization module is responsible for calculating the optimal dead time based on the estimated virtual junction temperature in the temperature state sub-vector and outputting it to the PWM generator dead time register.

12. The system according to claim 3, characterized in that, The active thermal stress management module executes two active thermal management strategies: nighttime insulation and high-temperature frequency modulation peak shaving, based on the virtual junction temperature estimate in the temperature state sub-vector, the current state of the intelligent protection state machine module in the system mode sub-vector, and the request flag set by the adaptive FNN-PID control module.

13. The system according to claim 3, characterized in that, The edge digital twin health management module performs predictive maintenance and intelligent health management based on the virtual junction temperature estimate in the temperature state sub-vector and the power calculation mode flag in the active thermal stress management module.

14. The system according to claim 13, characterized in that, The predictive maintenance and intelligent health management includes real-time simulation of reduced-order thermal models, operation of digital twin simulation engine, online correction of model parameters, rainflow counting and cumulative damage calculation, multi-failure mode damage fusion, prediction of health indicator degradation trends, prediction of remaining life, and predictive maintenance decision support.

15. The system according to claim 3, characterized in that, The impedance environment sensing module performs a complete impedance spectrum scan and updates the environment state subvector when the system is in standby or the load rate is below 10%. When the system is in normal operation and the load rate is consistently above 10%, the narrowband fast impedance detection mode is enabled.

16. The system according to claim 3, characterized in that, The intelligent protection state machine module manages the system's operating status and makes decisions on protection actions based on the following information: Temperature information in the temperature state subvector; The cumulative damage level, thermal resistance information, and comprehensive health index in the health status subvector; The environment state in the environment state subvector; The current control parameters in the control state subvector.

17. The system according to claim 16, characterized in that, The management of the system's operating status and the decision-making for protection actions include: Determine and execute job state transitions; Dust accumulation adaptive dynamic derating; Recovery from abnormal working conditions.

18. The system according to claim 3, characterized in that, The line voltage drop compensation module calculates the voltage drop compensation amount of the remote cable in real time based on the temperature information in the temperature state sub-vector and the load current obtained by the power level sampling circuit.

19. A control method for an adaptive wide-temperature-range power supply controller, characterized in that, The method is used to control the adaptive wide-temperature-range power controller system according to any one of claims 1-18, and the method includes system initialization, high-frequency real-time control, medium-frequency adaptive adjustment, low-frequency health management and environmental perception.

20. The method according to claim 19, characterized in that, The system initialization includes: Perform a self-test on the system hardware; Physical temperature is acquired using a thermistor sensor; The impedance environment sensing module performs rapid impedance self-test to quickly determine the environment category. The edge digital twin health management module is initialized to verify whether the deviation between the initial output of the digital twin and the physical measurement is within a reasonable threshold range; Make an initial working mode decision to determine the working mode of the module.

21. The method according to claim 20, characterized in that, The initial operating mode decision, to determine the operating mode of the module, includes: If the ambient temperature is lower than the set temperature or the environment is damp and dirty, the intelligent protection state machine module enters the preheating state; if the ambient temperature is not lower than the set temperature and the environment is not damp and dirty, the intelligent protection state machine module enters the soft start state; if the comprehensive health index loaded by the edge digital twin health management module is greater than the set value, then regardless of the temperature and environmental conditions, the intelligent protection state machine module enters the dynamic derating state to protect the severely degraded power module.

22. The method according to claim 19, characterized in that, The high-frequency real-time control includes the controller system cyclically executing the following high-frequency control functions after the intelligent protection state machine module enters the soft-start state or normal operation state: Synchronous data sampling: to collect data such as voltage, current, and on-state voltage drop of the controller; Virtual junction temperature estimation update: The junction temperature is calculated by substituting the on-state voltage drop sample value into the calibration equation, and then the virtual junction temperature estimation value is updated after 8-point moving average filtering. ESR Online Estimation Update: Based on the voltage AC ripple component and the inductor current ripple component, calculate and update the online ESR estimate; Observer iteration for multi-source information fusion: Perform a one-step iteration of the discrete state equations of the multi-source information fusion observer; PID control law calculation: including voltage error calculation, where the reference voltage is the value after compensation by the line voltage drop compensation module; Hardware protection monitoring: If the inductor current or output voltage exceeds the rated threshold range, the hardware comparator immediately triggers an interrupt and forces the intelligent protection state machine module to switch to the emergency cut-off state.

23. The method according to claim 19, characterized in that, The intermediate frequency adaptive adjustment includes the controller system cyclically performing the following intermediate frequency adaptive adjustment: Fuzzy inference based on temperature data is used to weighted synthesize PID adjustment coefficients; The final PID parameters are synthesized through a neural network, while backpropagation is performed to update the network weights. Mode arbitration and execution of the active thermal stress management module: The active thermal stress management module reads the request flag and the status of the intelligent protection state machine module, determines whether to accept the request according to the arbitration priority, and executes the corresponding mode switch; PID parameters are adjusted in conjunction with frequency modulation: If the current mode is frequency modulation mode of the active thermal stress management module, the PID parameters are scaled twice to maintain the phase margin at a value not lower than the set phase margin value. Dead time update: Read the estimated virtual junction temperature from the temperature state subvector, calculate and update the temperature-compensated dead time; Digital twin iteration: Perform a one-step iteration of the reduced-order state equation; Compensation calculation of line voltage drop compensation module: Calculate the voltage drop compensation amount of the far-end cable, and update and adjust the reference voltage after upper limit clamping for high frequency cycle use; Intelligent protection state machine module state transition judgment: Check whether the comprehensive health indicators exceed the warning threshold; if so, trigger the transition to dynamic depreciation state. Double-buffered pointer flip: Reverses the read / write roles of two equal-length memory regions through atomic operations.

24. The method according to claim 19, characterized in that, The low-frequency health management and environmental perception include the controller system performing the following low-frequency health management and environmental perception in a non-preemptive manner: Perform thermistor sensor sampling and thermal field reconstruction: Perform filtered sampling on the thermistor sensor, and substitute it into the cubic spline interpolation algorithm to calculate the 8×6 grid temperature distribution map; Perform thermal resistance trend analysis: calculate the dust accumulation degradation coefficient. If the change in the dust accumulation degradation coefficient exceeds the set dust accumulation degradation coefficient, dynamically adjust the derating start temperature and notify the intelligent protection state machine module. Perform rainflow counting and damage update: Perform rainflow counting on the junction temperature sequence of the most recent 10 seconds, extract the thermal cycle, calculate the theoretical lifetime, and update the cumulative damage. Perform multi-failure mode damage fusion: After completing the damage update in the previous step, calculate the bond line lift-off damage and package delamination damage, and then weight and fuse the bond line lift-off damage, package delamination damage, and solder layer fatigue damage to obtain a comprehensive health index. Hourly degradation trend fitting and remaining useful life (RUL) prediction: Read the historical sequence of the daily average value of the comprehensive health index for the past 30 days, use the weighted least squares method to fit the linear degradation trend, and extrapolate the nominal value, optimistic value and pessimistic value of remaining useful life based on the degradation slope. Impedance spectrum measurement of the impedance environment sensing module; Online model calibration of the edge digital twin health management module: Compare the digital twin output with the observer output every hour or after the operating condition changes. If the deviation exceeds the set deviation value for 10 consecutive times, start the recursive least squares incremental correction. Perform health log persistence every hour: Write a snapshot of the current health status to non-volatile storage.

25. The method according to claim 19, characterized in that, The method also includes state transition linkage control, which is an event-driven process triggered by the intelligent protection state machine module within the medium-frequency adaptive adjustment cycle.

26. The method according to claim 25, characterized in that, The aforementioned state transition linkage control includes the linkage process for the intelligent protection state machine module to transition from the normal operation state to the dynamic derating state, as well as the linkage process for recovery from the emergency disconnection state.

Images