Machine learning based method and system for predicting lifetime of LED driving circuit

Abstract

This invention belongs to the field of artificial intelligence and digital data processing technology, and relates to a method and system for predicting the lifetime of LED driver circuits based on machine learning. The method includes: acquiring the ambient temperature, operating voltage, and current sequences of the circuit board containing the core components of the LED driver circuit; extracting the time features of the temperature sequences to determine the cumulative value of dynamic temperature fluctuations; determining a dynamic thermal time constant correction factor based on the cumulative value; reconstructing the temperature sequence according to the correction factor to determine a virtual core temperature condition variable; and inputting the voltage and current sequences as the main input and the virtual core temperature as the condition variable into a pre-trained conditional variational autoencoder to output the remaining lifetime prediction result. This invention can more accurately assess the cumulative effect of ambient temperature fluctuations, scientifically correct the influence of thermal inertia reflected in the heat conduction process, effectively solve the problem of feature decoupling distortion, improve the accuracy and reliability of LED driver circuit lifetime prediction, reduce the risk of unexpected failures, and ensure stable circuit operation.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and digital data processing technology, and relates to a method and system for predicting the lifespan of LED driver circuits based on machine learning. Background Technology

[0002] The driver circuit of automotive LED headlights and taillights is a core component ensuring driving safety. Based on the "weakest link" principle, the overall lifespan of an LED driver circuit is determined by its most vulnerable core components, such as heavy-load electrolytic capacitors and high-power MOSFETs. Therefore, accurately predicting the remaining lifespan of these core components to obtain the overall lifespan index of the LED driver circuit has significant industrial value for preventative vehicle maintenance.

[0003] In conventional machine learning-based lifetime prediction techniques, electrical parameters such as operating voltage and current sequences of core components are typically collected and input into the prediction model. Due to the harsh operating environment of automobiles, to eliminate transient drift interference caused by temperature on electrical parameters, existing technologies often introduce conditional variational autoencoders (CDAEs) as prediction models. The ambient temperature of the circuit board, collected by sensors, is input into the CDAE as a synchronous conditional variable. This allows for the separation of temperature-induced electrical parameter fluctuations from the inherent health degradation characteristics of the components within the low-dimensional feature space constructed by the model, yielding temperature-independent features and thus enabling lifetime prediction.

[0004] However, in practical applications, the existing technology based on conditional variational autoencoders suffers from several drawbacks. The existing model implicitly assumes that changes in external temperature will synchronously alter the electrical parameters of core components. In real automotive LED driver circuits, when temperature jumps occur, such as engine overheating or cold water splashing, heat conduction to the core components exhibits significant thermal inertia and delay. Because the ambient temperature of the circuit board at the sensor measurement point is misaligned with the actual internal temperature of the core components in time, the model uses leading or lagging temperature conditions to decouple the current electrical parameters, leading to distorted feature decoupling. This results in the inability to extract accurate circuit health degradation features and introduces additional decoupling noise, ultimately affecting the accuracy of LED driver circuit lifetime prediction. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and solve the technical problem that the model timing input misalignment and feature decoupling distortion caused by the thermal inertia of core components ultimately affect the accuracy of LED driver circuit lifetime prediction. The invention provides a method and system for LED driver circuit lifetime prediction based on machine learning.

[0006] To achieve the above-mentioned objectives, this invention provides a method for predicting the lifetime of LED driver circuits based on machine learning, comprising the following steps: The process involves acquiring the circuit board ambient temperature sequence, operating voltage sequence, and operating current sequence of the core components in the LED driver circuit; extracting time series features based on the circuit board ambient temperature sequence to determine the cumulative dynamic temperature fluctuation value of the environment where the core components are located; determining the dynamic thermal time constant correction factor for the equivalent heat conduction process of the core components based on the cumulative dynamic temperature fluctuation value; reconstructing the circuit board ambient temperature sequence according to the dynamic thermal time constant correction factor to determine the virtual core temperature condition variable inside the core components; synchronously inputting the operating voltage sequence and operating current sequence as the main input, and the virtual core temperature condition variable as the condition variable, into a pre-trained conditional variational autoencoder to output the remaining service life prediction result of the core components, and using the remaining service life prediction result of the core components as the service life prediction result of the LED driver circuit.

[0007] This invention assesses ambient temperature changes by determining the cumulative value of dynamic temperature fluctuations, accurately reflecting the cumulative effect of temperature fluctuations and providing a reliable temperature characteristic basis for subsequent calculation of dynamic thermal time constant correction factors. By determining the dynamic thermal time constant correction factor, it achieves dynamic correction of the heat conduction process, accurately reflecting the influence of thermal inertia on temperature transfer and providing a reliable basis for the time-series reconstruction of circuit board ambient temperature sequences. By determining virtual core temperature condition variables through time-series reconstruction, it achieves accurate estimation of the internal temperature of core components, effectively solving the time misalignment problem between external and internal temperatures and providing accurate temperature condition input for conditional variational autoencoders. Based on the lifetime prediction of virtual core temperature condition variables, it effectively solves the feature decoupling distortion problem, improving the accuracy and reliability of LED driver circuit lifetime prediction.

[0008] Furthermore, the circuit board ambient temperature sequence, operating voltage sequence, and operating current sequence of the core components in the LED driver circuit are obtained, including: acquiring the circuit board ambient temperature sequence of the core components in a continuous time series using an ambient temperature sensor arranged on the edge of the LED driver circuit board at a fixed sampling period; synchronously acquiring the operating voltage sequence at the corresponding time node using a voltage detection sampling circuit at a fixed sampling period; and synchronously acquiring the operating current sequence at the corresponding time node using a current detection sampling circuit at a fixed sampling period.

[0009] Furthermore, the cumulative dynamic temperature fluctuation of the environment in which the core components are located is determined, including: In the formula, For the current moment The cumulative value of dynamic temperature fluctuations in the environment where the core components are located. This represents the total number of sampling points within the historical observation window. This is the index of the backtracking steps for the time series. The current moment in the circuit board ambient temperature sequence The ambient temperature of the circuit board. The current moment in the circuit board ambient temperature sequence Backtracking The ambient temperature of the circuit board at each step. It is a natural exponential function.

[0010] This invention achieves a scientific assessment of the cumulative value of dynamic temperature fluctuations by constructing a sliding window summation model that includes an exponential decay weight for temperature difference. This model more accurately reflects the cumulative effect of historical temperature fluctuations, and the exponential decay weight ensures that recent temperature changes contribute more to the cumulative value, thus effectively characterizing the degree of cumulative impact of environmental temperature fluctuations.

[0011] Furthermore, the dynamic thermal time constant correction factor for the equivalent heat conduction process of the core components is determined, including: In the formula, For the current moment The dynamic thermal time constant correction factor for the equivalent heat conduction process of core components. The empirical weighting coefficients for core components, For the current moment The cumulative value of dynamic temperature fluctuations in the environment where the core components are located. For reference temperature constant, It is the hyperbolic tangent function.

[0012] This invention achieves a scientific evaluation of the dynamic thermal time constant correction factor by constructing a hyperbolic tangent function that includes the ratio of the cumulative value of dynamic temperature fluctuations to the reference temperature. This ensures that the correction factor varies around 1. The hyperbolic tangent function performs nonlinear normalization on the cumulative value of temperature fluctuations, thereby effectively reflecting the response characteristics of the heat conduction process to the cumulative value of temperature fluctuations.

[0013] Furthermore, the virtual core temperature condition variables within the core components are determined, including: In the formula, The virtual core temperature condition variable inside the core component at the current moment The value, The virtual core temperature condition variable inside the core component at the current moment The value at the previous moment, For a fixed sampling period, The nominal thermal time constant of the core components, For the current moment The dynamic thermal time constant correction factor for the equivalent heat conduction process of core components.

[0014] This invention achieves a scientific evaluation of virtual core temperature condition variables by constructing a first-order hysteresis filter model that includes a dynamic thermal time constant correction factor. This model more accurately reflects the time delay characteristics of the heat conduction process. The dynamic correction factor adaptively adjusts the thermal time constant according to ambient temperature fluctuations, thereby effectively simulating the real temperature response inside the core components and solving the problem of time misalignment between external and internal temperatures.

[0015] Furthermore, the value of the fixed sampling period must satisfy... .

[0016] Furthermore, before obtaining the circuit board ambient temperature sequence, operating voltage sequence, and operating current sequence of the core components in the LED driver circuit, the method further includes: during the initial power-on initialization, setting the value of the virtual core temperature condition variable of the core components at the initial moment to be equal to the circuit board ambient temperature at the same moment.

[0017] Furthermore, the remaining service life prediction results of the core components are output, including: inputting the operating voltage sequence, operating current sequence, and virtual core temperature condition variable into the encoder network of the conditional variational autoencoder, and stripping the transient drift features caused by temperature changes in the latent space through the encoder network to output a latent vector carrying the circuit health degradation features; inputting the latent vector into the regression network of the conditional variational autoencoder, and outputting the remaining service life prediction results of the core components based on the latent vector through the regression network.

[0018] This invention uses an encoder network to strip transient drift features from the latent space, achieving accurate extraction of health degradation features. This effectively eliminates electrical parameter fluctuation interference caused by temperature changes, ensuring that the latent vector accurately reflects the health degradation state of the circuit. By using a regression network to output the remaining service life prediction result based on the latent vector, the invention achieves accurate prediction of the lifespan of core components, improving the accuracy and reliability of LED driver circuit lifespan prediction.

[0019] Furthermore, it also includes: in response to the presence of multiple core components with different thermal characteristics in the LED driver circuit, for each core component, calculating independent virtual core temperature condition variables based on the corresponding nominal thermal time constant, and inputting the independent virtual core temperature condition variables as multi-channel condition variables into a pre-trained conditional variational autoencoder, outputting the remaining service life prediction results of multiple core components, and taking the minimum value among the remaining service life prediction results of multiple core components as the life prediction result of the LED driver circuit.

[0020] This invention also provides a machine learning-based LED driver circuit lifetime prediction system, which adopts the following technical solution: The LED driver circuit lifetime prediction system based on machine learning includes a processor and a memory. The memory stores computer program instructions, which are executed by the processor to implement the aforementioned LED driver circuit lifetime prediction method based on machine learning.

[0021] By adopting the above technical solution, the above-mentioned machine learning-based LED driver circuit lifetime prediction method is generated into a computer program and stored in a memory for loading and execution by a processor. This allows for the creation of terminal devices based on the memory and processor, making them convenient to use.

[0022] Compared with the prior art, the present invention has at least the following beneficial effects: Firstly, existing variational autoencoders implicitly assume that changes in ambient temperature and electrical parameters are synchronized, which cannot cope with thermal inertia and thermal delay during temperature jumps such as engine overheating or cold water splashes. This leads to timing misalignment and decoupling distortion between the external temperature of the sensor and the internal temperature of the components. This invention calculates the cumulative value of dynamic temperature fluctuations based on the circuit board ambient temperature sequence, derives the dynamic thermal time constant correction factor for the equivalent heat conduction process, and reconstructs the circuit board ambient temperature sequence in time. This generates a virtual core temperature condition variable that is completely synchronized with the changes in the electrical parameters of the core components. This effectively solves the decoupling errors caused by temperature conditions being ahead or behind, avoids the introduction of additional decoupling noise, and lays the foundation for the accurate extraction of health degradation characteristics.

[0023] Secondly, this invention uses the virtual core temperature after time-series reconstruction as a conditional variable input to the conditional variational autoencoder, replacing the circuit board ambient temperature directly used in traditional methods. This allows the model to accurately separate the transient drift of electrical parameters caused by temperature disturbances from the irreversible health degradation characteristics of the components themselves in a low-dimensional feature space. Compared with the decoupling failure and degradation characteristics masked by noise caused by temperature time-series misalignment in traditional methods, the health degradation characteristics extracted by this invention can better reflect the true aging state of the components, improving the accuracy and reliability of remaining service life prediction.

[0024] Third, considering the harsh operating environment of automobiles and the frequent and drastic temperature jumps, the dynamic thermal time constant correction factor of this invention can be adaptively adjusted according to the cumulative value of real-time temperature fluctuations. It does not require pre-calibration of fixed thermal time constants under different temperature jumps. It can autonomously adapt to various extreme temperature conditions such as sudden heating of the engine compartment, splashing of cold water in rainy weather, and low-temperature start-up in winter. It effectively solves the defect of the traditional method in the sharp drop in prediction accuracy under extreme temperature conditions and meets the life prediction requirements of vehicle electronic equipment under all operating conditions. Attached Figure Description

[0025] Figure 1 This is a flowchart of the LED driver circuit lifetime prediction method based on machine learning in an embodiment of the present invention.

[0026] Figure 2 This is a time-series comparison diagram of the circuit board ambient temperature and the virtual core temperature in the LED driver circuit lifetime prediction method based on machine learning in an embodiment of the present invention.

[0027] Figure 3 This is a schematic diagram illustrating the prediction results of the remaining service life of the core components of the LED driver in the LED driver circuit lifetime prediction method based on machine learning in an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings.

[0029] Example 1: This invention discloses a machine learning-based method for predicting the lifetime of LED driver circuits, with reference to... Figure 1 This includes steps S01-S05: S01: Obtain the circuit board ambient temperature sequence, operating voltage sequence, and operating current sequence of the core components in the LED driver circuit.

[0030] Specifically, an ambient temperature sensor is placed on the edge of the LED driver circuit board to fix the sampling period. Collect the circuit board ambient temperature sequence of core components over a continuous time series; use voltage and current detection sampling circuits with the same fixed sampling period. The operating voltage and operating current sequences at the corresponding time points are collected synchronously. At the same time, during the initial power-on initialization, the virtual core temperature condition variable of the core components is set to the value of the circuit board ambient temperature at the initial moment.

[0031] In this embodiment, a fixed sampling period is used. The value of must satisfy , The nominal thermal time constant of the core component is obtained from the datasheet of the core component, such as a specific model of electrolytic capacitor or MOSFET, or through standard thermal testing, such as JEDEC standard measurements. This constraint is based on the fact that the virtual core temperature calculation formula in subsequent steps is constructed based on a first-order heat conduction model, and the sampling period... Much larger than the thermal time constant The weighting coefficients in the formula This will approach 1, resulting in a calculated virtual core temperature. convergence with circuit board ambient temperature Thus, the thermal time constant is lost. The thermal inertia hysteresis characteristic it represents renders the thermal inertia compensation mechanism ineffective. The choice of 0.1 is based on the engineering experience of balancing numerical calculation accuracy and system resource consumption. According to the general engineering experience rules in the field of first-order system discretization, at least 10 sampling points are required within each time constant, that is, the step size is 0.1 times the time constant, in order to control the discretization truncation error within 5% and thus better fit the exponential change curve. This value is to avoid unnecessary increase in processor computing load caused by too small a sampling period, while ensuring that the thermal inertia characteristics are not distorted.

[0032] S02: Extract time series features based on the circuit board ambient temperature sequence to determine the cumulative value of dynamic temperature fluctuations in the environment where the core components are located.

[0033] It should be noted that the thermal hysteresis of core components is not a constant value; it is directly related to the intensity, duration, and cumulative effect of recent ambient temperature fluctuations. A single nominal thermal time constant cannot accurately characterize the actual thermal response characteristics of the device under different thermal shock conditions. Therefore, this step extracts the temporal fluctuation characteristics of ambient temperature and introduces an exponentially decaying weight for sliding window accumulation to characterize the comprehensive intensity of the external thermal shock experienced by the device at the current moment, providing a basis for subsequent dynamic correction of the thermal time constant.

[0034] Specifically, determining the cumulative dynamic temperature fluctuation of the environment in which the core components are located includes: ; In the formula, For the current moment The cumulative value of dynamic temperature fluctuations in the environment where the core components are located. The total number of sampling points in the historical observation window. In this embodiment, the length of the historical observation window is... The solid heat conduction process can be approximated as a first-order inertial system. Taking this length as the observation window can fully capture all the significant thermal response processes of the components after a temperature step, without missing the cumulative effect of thermal shock or including historical data that no longer has a substantial impact. This is the index of the backtracking steps for the time series. The current moment in the circuit board ambient temperature sequence The ambient temperature of the circuit board. The current moment in the circuit board ambient temperature sequence Backtracking The ambient temperature of the circuit board at each step. It is a natural exponential function.

[0035] The absolute value of the temperature difference represents the fluctuation range of the ambient temperature from the historical time to the current time. Multiplying it by the exponential decay weight and accumulating it within a window objectively reflects the cumulative effect of external thermal shock on the device surface during that period. When the ambient temperature is in a condition of drastic step change and continuous fluctuation, the absolute value of the temperature difference corresponding to the recent backtracking step increases and the weight is close to 1. The value of the dynamic temperature fluctuation accumulation index increases, which objectively represents the severity of the thermal shock suffered by the device and provides a basis for subsequent correction of the thermal time constant.

[0036] The above-mentioned relationship is based on the first-order inertial system theory of solid heat conduction in physics. It aims to scientifically assess the cumulative dynamic temperature fluctuations of a circuit board's environment, more accurately reflecting the cumulative effect of historical temperature fluctuations on core components. Traditional lifetime prediction methods often neglect the nonlinear correlation between the degree of thermal shock to core components and the intensity and duration of recent environmental temperature fluctuations. Since the impact of historical thermal shocks on the current thermodynamic state of an object is not constant but rather exhibits exponential energy dissipation and decay over time, this invention introduces a natural exponential function based on sliding window summation. As a decay weight over time, the absolute value of the temperature difference represents the fluctuation amplitude. Combined with the exponential penalty, it conforms to the physical law of thermal response memory in the natural environment, so that the recent temperature change closer to the current moment contributes more to the cumulative value, and more accurately assesses the intensity of the real cumulative effect of external thermal shock on the device surface during this period.

[0037] The temperature difference term in the above relationship The dimension is temperature, and the weight is exponentially decaying. Since the two are dimensionless pure numbers, their product and summation via a sliding window result in a total value of 1. The dimension of is still temperature, representing the weighted cumulative index of temperature fluctuation.

[0038] S03: Based on the cumulative value of dynamic temperature fluctuations, determine the dynamic thermal time constant correction factor for the equivalent heat conduction process of core components.

[0039] It should be noted that when a severe thermal shock occurs in the external environment, the heat absorption rate of the device's casing material and the interfacial thermal resistance between the casing and the internal chip will exhibit nonlinear changes, causing the actual thermal time constant of the device to deviate from the factory nominal value. Therefore, this step generates a dynamic correction factor based on the cumulative intensity of thermal shock through a nonlinear mapping relationship that conforms to the physical laws of heat conduction, thereby achieving real-time adaptive adjustment of the nominal thermal time constant.

[0040] Specifically, the dynamic thermal time constant correction factor for the equivalent heat conduction process of core components is determined, including: ; In the formula, For the current moment The dynamic thermal time constant correction factor for the equivalent heat conduction process of core components. The empirical weighting coefficients for core components are obtained through a validation set grid search during the offline model training phase, with the goal of minimizing lifetime prediction error. In this embodiment... The range of values ​​is ; For the current moment The cumulative value of dynamic temperature fluctuations in the environment where the core components are located. As a reference temperature constant, in this embodiment Used to remove The dimensions of _____ are different. During normal car operation, the temperature drift in the natural environment is slow, typically much less than 5°C per minute. Only when external forced thermal shocks occur, such as sudden high-load overheating of the engine or splashing cold water onto the chassis drive plate, will temperatures exceeding _____ accumulate in a short period. Therefore, this value serves as the physical critical threshold for distinguishing between temperature drift in normal environments and abnormal thermal shocks under automotive-grade operating conditions. It ensures that the modified model can quickly trigger nonlinear thermal hysteresis compensation when facing extreme thermal shocks, while avoiding calculation oscillations in life prediction results due to oversensitivity under steady-state operating conditions. It is the hyperbolic tangent function.

[0041] Specifically, by constructing a nonlinear mapping relationship based on the hyperbolic tangent function, the above relationship conforms to the natural physical law that when the external thermal shock accumulates to a certain extent, the heat absorption of the device shell material reaches saturation, resulting in a nonlinear extension of the hysteresis time for heat conduction into the interior. This transforms the quantifiable temperature fluctuation index into a dimensionless expansion factor used to correct the thermal time constant. The reference temperature constant serves as a dimensional conversion bridge, ensuring that the independent variable of the hyperbolic tangent function is a pure number, thus avoiding errors in mathematical definition.

[0042] The above relationship is based on the physical laws of material heat capacity and thermal resistance, aiming to generate a dynamic correction factor based on the cumulative intensity of thermal shock, and to achieve real-time adaptive adjustment of the nominal thermal time constant of core components. When the external environment experiences severe thermal shocks such as engine overheating or cold water splashing, the heat absorption rate of the device shell material and the interfacial thermal resistance between the shell and the internal chip will change nonlinearly. In order to conform to the physical law that when the external thermal shock accumulates to a certain extent, the heat absorption of the device shell material reaches saturation, resulting in a nonlinear increase in the hysteresis time of heat conduction to the interior, this invention utilizes a hyperbolic tangent function that can converge to a physical limit value. A nonlinear mapping relationship was constructed; a reference temperature constant was introduced. Remove The dimension of the quantifiable historical temperature fluctuation index is scientifically mapped to the dimensionless thermal time constant expansion factor.

[0043] The cumulative value of dynamic temperature fluctuation in the independent variable of the above relationship With reference temperature constant Both are dimensionless (temperature), and dividing them cancels out their dimensions, transforming them into dimensionless pure numbers; based on this, after... Operations and dimensionless empirical weighting coefficients Multiply and add 1 to get the final correction factor. It is a dimensionless multiplier constant.

[0044] S04: Based on the dynamic thermal time constant correction factor, the circuit board ambient temperature sequence is reconstructed in time to determine the virtual core temperature condition variables inside the core components.

[0045] It should be noted that directly embedding a temperature sensor inside the core component would damage the original structure of the device, increase hardware costs and process complexity, and would not be compatible with standardized drive circuits for mass production. Therefore, this step uses a modified dynamic thermal time constant and a first-order thermal conduction numerical model to reconstruct the timing of the circuit board ambient temperature, indirectly obtaining a virtual core temperature synchronized with changes in the device's electrical parameters.

[0046] Specifically, the virtual core temperature condition variables inside the core components are determined, including: ; In the formula, The virtual core temperature condition variable inside the core component at the current moment The value, The virtual core temperature condition variable inside the core component at the current moment The value at the previous moment, For a fixed sampling period, The nominal thermal time constant of the core components, For the current moment The dynamic thermal time constant correction factor for the equivalent heat conduction process of core components.

[0047] Specifically, a dynamic thermal time constant correction factor is used to dynamically amplify the time constant term of the heat conduction equation. Under conditions where drastic changes in external temperature cause thermal inertia misalignment, the increase in the dynamic thermal time constant correction factor leads to an increase in the denominator value. In the temperature iteration calculation, these changes enable the system to automatically compress the change in virtual core temperature compared to the previous moment, causing the reconstruction process to rely more on the physical laws of thermal inertia rather than instantaneous ambient temperature during harsh environments. This eliminates the time series misalignment between the sensor measurement point temperature and the actual internal temperature of the device, ensuring that the conditional variables input to the conditional variational autoencoder can accurately remove transient drift caused by changes in real temperature.

[0048] The above relationship is derived from the first-order thermal conduction numerical model describing heat transfer in solids in physics, aiming to indirectly and accurately derive the virtual internal temperature synchronized with changes in the electrical parameters of core components. Existing low-pass filtering or first-order thermal conduction models use a fixed thermal time constant, implicitly assuming the synchronization of internal and external temperature changes. This invention utilizes a dynamic correction factor obtained from prior calculations in the classical iterative formula. For the original nominal thermal time constant Dynamic scaling and replacement were performed, replacing with When faced with harsh working conditions involving drastic temperature changes, It will actively increase and drive the denominator value to rise, so that the system automatically compresses the change in virtual core temperature compared to the previous moment. This ensures that the reconstruction process strictly follows the thermodynamic conservation and natural conduction laws that heat transfer cannot be completed instantaneously and must exhibit an exponentially smooth transition curve. This effectively eliminates the time misalignment between the external sensor temperature and the internal real temperature.

[0049] The above relationship has a fixed sampling period. With the nominal thermal time constant The dimension of all terms is time. In the fractional terms... In the middle, because Since the term is dimensionless, the denominator still has the dimension of time. The dimensions of the numerator and denominator cancel each other out, resulting in a dimensionless smoothing weighting coefficient. This smoothing weighting coefficient is multiplied by the temperature difference term, which has the dimension of temperature, and then summed to finally derive the virtual core temperature. Temperature dimensions were strictly maintained.

[0050] like Figure 2 As shown in the figure, the horizontal axis represents time in seconds (s), and the vertical axis represents temperature in degrees Celsius (°C). It can be seen from the graph that... At that moment, a sudden temperature jump occurred in the engine, with the ambient temperature at the measuring point instantly rising from about 25°C to about 40°C. The corrected virtual core temperature exhibited an exponential increase curve that conformed to the first-order heat conduction law, accurately reproducing the temperature delay response characteristics caused by thermal inertia inside the core component. This verified that the timing reconstruction method of the present invention can effectively eliminate the time series misalignment problem between the external temperature of the sensor and the internal temperature of the component.

[0051] S05: The operating voltage sequence and operating current sequence are used as the main inputs, and the virtual core temperature condition variable is used as the condition variable and synchronously input into the pre-trained conditional variational autoencoder. The remaining service life prediction result of the core components is output, and the remaining service life prediction result of the core components is used as the service life prediction result of the LED driver circuit.

[0052] It should be noted that the operating voltage and current signals of the LED driver circuit simultaneously contain reversible transient drift components caused by temperature changes and irreversible degradation components caused by component material aging. These two types of components are superimposed and coupled in the original signal space. Therefore, this step adopts a conditional variational autoencoder architecture, using the virtual core temperature as a conditional variable, to achieve accurate decoupling of the two types of components in a low-dimensional latent space, and then to predict the remaining service life based on pure degradation characteristics.

[0053] Specifically, the predicted remaining service life of the core components is output, including: The operating voltage sequence, operating current sequence, and virtual core temperature condition variable are input into the encoder network of the conditional variational autoencoder. The encoder network strips the transient drift features caused by temperature changes in the latent space and outputs a latent vector carrying the characteristics of circuit health degradation. The latent vector is input into the regression network of the conditional variational autoencoder, and the regression network outputs the remaining service life prediction result of the core component based on the latent vector. In response to the presence of various core components with different thermal characteristics in the LED driver circuit, an independent virtual core temperature condition variable is calculated for each core component based on its corresponding nominal thermal time constant. This independent virtual core temperature condition variable is then input as a multi-channel condition variable into a pre-trained conditional variational autoencoder, which outputs the remaining service life prediction results for multiple core components. The minimum value among the remaining service life prediction results for multiple core components is then used as the lifespan prediction result for the LED driver circuit.

[0054] like Figure 3As shown, the horizontal axis represents time in seconds, and the vertical axis represents the predicted remaining service life in hours. It can be seen that the prediction curve shows a smooth linear downward trend, without any oscillation in the predicted value caused by temperature steps. This proves that by introducing a virtual core temperature condition variable, the present invention successfully stripped the transient temperature drift features in the latent space of the conditional variational autoencoder, accurately extracted the irreversible health degradation features of the components, and achieved stable and accurate prediction of the remaining service life.

[0055] This invention also discloses a machine learning-based LED driver circuit lifetime prediction system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the machine learning-based LED driver circuit lifetime prediction method according to this invention.

[0056] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for predicting the lifetime of LED driver circuits based on machine learning, characterized in that, include: Obtain the circuit board ambient temperature sequence, operating voltage sequence, and operating current sequence of the core components in the LED driver circuit; Based on the circuit board ambient temperature sequence, time series features are extracted to determine the cumulative value of dynamic temperature fluctuations in the environment where the core components are located. Based on the cumulative value of dynamic temperature fluctuations, the dynamic thermal time constant correction factor for the equivalent heat conduction process of core components is determined. The circuit board ambient temperature sequence is reconstructed based on the dynamic thermal time constant correction factor to determine the virtual core temperature condition variables inside the core components. The operating voltage sequence and operating current sequence are used as the main inputs, and the virtual core temperature condition variable is used as the condition variable and synchronously input into the pre-trained conditional variational autoencoder. The output is the prediction result of the remaining service life of the core components, and the prediction result of the remaining service life of the core components is used as the prediction result of the life of the LED driver circuit.

2. The method for predicting the lifetime of LED driver circuits based on machine learning according to claim 1, characterized in that, Obtain the circuit board ambient temperature sequence, operating voltage sequence, and operating current sequence of the core components in the LED driver circuit, including: An ambient temperature sensor is placed on the edge of the LED driver circuit board to collect the circuit board ambient temperature sequence of the core components in a continuous time series at a fixed sampling period. The voltage detection and sampling circuit synchronously acquires the operating voltage sequence at the corresponding time node at a fixed sampling period; The current detection and sampling circuit synchronously collects the operating current sequence at the corresponding time node at a fixed sampling period.

3. The method for predicting the lifetime of LED driver circuits based on machine learning according to claim 1, characterized in that, Determine the cumulative dynamic temperature fluctuation of the environment in which the core components are located, including: ； In the formula, For the current moment The cumulative value of dynamic temperature fluctuations in the environment where the core components are located. This represents the total number of sampling points within the historical observation window. This is the index of the backtracking steps for the time series. The current moment in the circuit board ambient temperature sequence The ambient temperature of the circuit board. The current moment in the circuit board ambient temperature sequence Backtracking The ambient temperature of the circuit board at each step. It is a natural exponential function.

4. The method for predicting the lifetime of LED driver circuits based on machine learning according to claim 1, characterized in that, Determine the dynamic thermal time constant correction factor for the equivalent heat conduction process of core components, including: ； In the formula, For the current moment The dynamic thermal time constant correction factor for the equivalent heat conduction process of core components. The empirical weighting coefficients for core components, For the current moment The cumulative value of dynamic temperature fluctuations in the environment where the core components are located. As a reference temperature constant, It is the hyperbolic tangent function.

5. The method for predicting the lifetime of LED driver circuits based on machine learning according to claim 1, characterized in that, Determine the virtual core temperature condition variables inside the core components, including: ； In the formula, The virtual core temperature condition variable inside the core component at the current moment The value, The virtual core temperature condition variable inside the core component at the current moment The value at the previous moment, For a fixed sampling period, The nominal thermal time constant of the core components, For the current moment The dynamic thermal time constant correction factor for the equivalent heat conduction process of core components.

6. The method for predicting the lifetime of LED driver circuits based on machine learning according to claim 5, characterized in that, The value of the fixed sampling period must satisfy the following: .

7. The method for predicting the lifetime of LED driver circuits based on machine learning according to claim 5, characterized in that, Before obtaining the circuit board ambient temperature sequence, operating voltage sequence, and operating current sequence of the core components in the LED driver circuit, the following steps are also included: During the initial power-on initialization, the virtual core temperature condition variable of the core components is set to the value of the circuit board ambient temperature at the same time.

8. The method for predicting the lifetime of LED driver circuits based on machine learning according to claim 1, characterized in that, The remaining service life prediction results for the core components are output, including: The operating voltage sequence, operating current sequence, and virtual core temperature condition variable are input into the encoder network of the conditional variational autoencoder. The encoder network strips the transient drift features caused by temperature changes in the latent space and outputs a latent vector carrying the characteristics of circuit health degradation. The latent vector is input into the regression network of the conditional variational autoencoder, and the regression network outputs the remaining service life prediction result of the core component based on the latent vector.

9. The method for predicting the lifetime of LED driver circuits based on machine learning according to claim 5, characterized in that, Also includes: In response to the presence of various core components with different thermal characteristics in the LED driver circuit, an independent virtual core temperature condition variable is calculated for each core component based on its corresponding nominal thermal time constant. This independent virtual core temperature condition variable is then input as a multi-channel condition variable into a pre-trained conditional variational autoencoder, which outputs the remaining service life prediction results for multiple core components. The minimum value among the remaining service life prediction results for multiple core components is then used as the lifespan prediction result for the LED driver circuit.

10. A machine learning-based LED driver circuit lifetime prediction system, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the machine learning-based LED driver circuit lifetime prediction method according to any one of claims 1 to 9.

Images