Lifetime-extending method by means of temperature regulation of an electronic component based on a load prediction

Abstract

The invention relates to a computer-implemented method comprising the following aspects: the load controller transmits current temperature measurement data of the electronic component to the temperature regulation and to the temperature simulation; current load measurement data and temperature measurement data of the electronic component are also transmitted to the AI-based load prediction and to the AI training. Priority is given to extending the lifetime rather than to immediate efficiency maximisation, since lifetime has become a critical design criterion throughout the electronics sector. For this purpose, the temperature of the electronic component is regulated by means of the load controller and optionally by active thermal management.

Description

[0001] Description

[0002] Lifespan-enhancing method using temperature regulation of an electronic component based on a load forecast

[0003] The present invention relates to a computer-implemented method and arrangement for carrying out the method for controlling an electronic component.

[0004] State of the art

[0005] In the field of electrical systems, particularly electronic components, the variability of power requirements presents an immense technical challenge. Electronic devices are subjected to thermal stress, which causes material fatigue. This phenomenon is primarily due to the constant changes in power demand. These fluctuations trigger numerous temperature cycles throughout the device's lifespan. Each temperature cycle induces thermomechanical stress. Due to differing coefficients of thermal expansion, mechanical stress arises specifically at transition layers (CTE mismatch). This thermomechanical stress affects not only semiconductor dies but also various components within the electronic modules.These components exhibit different coefficients of thermal expansion, resulting in varying responses to thermal stresses. Consequently, the cumulative effect of these stresses, caused by repeated temperature cycles, can lead to accelerated degradation and eventual failure of the interface to the electronic module. This problem is exacerbated by unpredictable fluctuations in power demand, posing a significant challenge to the effective management and reduction of the associated thermal stress in order to improve the longevity and reliability of electronic devices.

[0006] However, thermal fatigue is not limited to power electronic devices, but affects various electrical systems, including batteries, general printed circuit boards (PCBs), including central processing units (CPUs) and graphics processing units (GPUs), or electric motors.

[0007] Consequently, it is technically necessary to propose an improved solution that avoids the disadvantages known from the prior art. In particular, the proposed solution should reduce or delay thermal fatigue of the electronic component. Description of the invention

[0008] The computer-implemented method according to the invention comprises temperature control of an electronic component, in particular a power electronic component. For temperature control, a load forecast for the electronic component is performed, which is then incorporated into a temperature simulation. Using this temperature simulation, a temperature prediction for the electronic component is determined over a specific period, incorporating load measurement data and temperature measurement data from the electronic component. The load forecast is generated using computer logic (CL). Based on the load forecast and temperature prediction, the temperature of the electronic component is regulated by means of load control.

[0009] The invention strategically utilizes electrical energy to regulate temperature levels under certain conditions, thereby advantageously extending the overall lifespan of electronic devices.

[0010] This approach prioritizes extending lifetime over immediate efficiency maximization, a departure from conventional design and operating practices that typically aim to minimize energy losses. Lifespan is a critical design criterion throughout the entire power electronics system architecture. The selection of power modules, cooling chains, and assembly technologies is based on defined load cycles and expected lifetimes. The challenge lies particularly in differentiating between varying application requirements, such as those found in pump, fan, and crane applications. The same power unit used in these applications will exhibit different lifetime behaviors.

[0011] In an advantageous variant of the computer-implemented method according to the invention, thermal as well as electrical performance parameters of the electronic component are controlled by means of the load control, based on setpoints received from the temperature control.

[0012] In particular, load control includes adjusting performance parameters such as reactive power, varying a switching frequency, or modulating a switching speed.

[0013] Preferably, values, in particular operating limits, are stored in a system control unit of the electronic component and communicated to the temperature control unit.

[0014] In particular, further data are provided for temperature control, for example, electricity tariff prices, tariff forecasts and / or user configurations. In a further advantageous embodiment of the computer-implemented method according to the invention, the temperature measurement data included in the temperature simulation comprise at least the temperature of one component of the electronic component, in particular the temperature of a cooling component.

[0015] A cooling component is essentially any device designed to dissipate heat. This includes passive cooling devices such as heat sinks and active cooling methods such as coolant pumps. In the case of fans, direct temperature measurement is not possible due to convection-based heat transfer. Therefore, an alternative temperature value, particularly that of an air-cooled component, could be used. Inlet temperature or hotspot temperature can be determined, depending on the component of interest.

[0016] The proposed strategy aims to extend the lifespan of power electronic devices by intentionally controlling energy consumption to improve the longevity of the devices in various applications.

[0017] In the variant of the computer-implemented method according to the invention described above, temperature values ​​of further individual components of the electronic component, in particular individual IGBTs and / or diodes, are preferably calculated using a thermal model.

[0018] For example, in the proposed computer-implemented method, the thermal model is derived from a 3D thermal simulation, which takes into account cross-component couplings using model reduction.

[0019] In one of the two variants of the computer-implemented method described above, an individual temperature profile can preferably be generated based on the load prediction for several components within the electronic component.

[0020] Figure 4 shows such a temperature profile, for example. This can then be used to decide which troughs should be bridged by additional heating in order to extend the service life.

[0021] In a particularly advantageous embodiment of the computer-implemented method according to the invention, the temperature of the electronic component is regulated by means of active thermal management. Thermal management can, for example, be understood as a cooling process. However, there are also scenarios in which a power electronic component is deliberately heated.

[0022] Active thermal management preferably includes a cooling process, which includes controlling a liquid pump flow, controlling a fan speed, regulating reactive power, regulating energy output and / or applying a distributable load to fine-tune the cooling capability.

[0023] The control process therefore preferably consists of two steps: Load control adjusts the operating sequences based on the generated load forecast. Predefined setpoints are applied. These setpoints are received, in particular, by the temperature control system. This load control manages the electrical power parameters of the electronic component and the cooling system, which in turn control the thermal parameters. In addition to load control, temperature regulation takes place, whereby the optimized load then enables active management of the thermal system. Temperature regulation can be achieved by controlling passive cooling systems as well as active cooling systems, especially solid-state coolers, such as thermoelectric cooling systems.

[0024] In one of the two variants of the computer-implemented method according to the invention described above, a two-criteria problem is solved, in particular one involving temperature fluctuation minimization and energy efficiency or energy availability. This is solved specifically using a computer algorithm. The active thermal management is then tailored to the cooling capacity of the cooling system used in the electronic component.

[0025] This approach addresses the challenge of determining the optimal degree of intervention in the cooling curve to minimize temperature fluctuations. This calculation requires load forecasting as an input. Accordingly, the approach complements the bi-criteria optimization problem: a suitable weighting factor must be chosen between the two criteria to calculate the optimum. Energy availability can also be made dependent on energy costs or electricity prices, for example.

[0026] Particularly advantageous in one embodiment of the method according to the invention is the input of data into the AI-based load forecast, in particular a data input that includes a data history.This data input can include, for example: various actual values, such as measurement data, especially actual load values ​​of an inverter; control input for temperature compensation; date and / or time, important for time-specific load cycles of inverters; other application-specific load parameters, such as weather forecasts, sun position, and similar data; other application-specific load parameters, such as whether the current day is a working day; planned load schedules of consumers or power sources in an island grid; a traffic forecast; a weather forecast; a production plan, especially for the following hours; and generally, a value history, especially a value history created by the process itself, for example, through the continuously updated input data. The input must be selected appropriately depending on the application.In a further preferred embodiment of the computer-implemented method according to the invention, measurement data from the electronic component are input into the AI-based load prediction. It is particularly advantageous that the load prediction AI is continuously trained during the execution of the method.

[0027] In the variant of the computer-implemented method according to the invention described above, for example, load measurement data and / or temperature measurement data of the electronic component are also input into the AI ​​training.

[0028] In particular, supervised learning (SL) is used in AI training. Alternatively, an alternative category of machine learning is applied, such as unsupervised learning (UL) or reinforcement learning (RL).

[0029] While the benefit of the proposed approach certainly depends on the type of application, investment costs, energy costs, and the CC footprint of the energy consumed, it will become increasingly important with regard to sustainability, dynamic electricity prices, and periods of surplus electricity from renewable energy sources. For example, battery conditioning—that is, preheating batteries before intensive charging and discharging—is already a common practice with proven advantages for lithium-ion batteries.

[0030] The proposed solution therefore offers broader applicability and addresses the problem of thermal fatigue in these systems. This extension is crucial because, like power electronic devices, these systems are subject to similar thermal stresses due to fluctuating power demands. By applying the principles and solutions developed for power electronic devices to these other systems, the invention paves the way for a more comprehensive approach to combating thermal fatigue.

[0031] The arrangement according to the invention for carrying out the computer-implemented method described above comprises a hardware component, in particular an electronic component with load control and temperature regulation, and a software component, in particular with temperature simulation and AI-based load prediction. The software component can, in particular, reside on or be executable on an edge device. Load control and temperature regulation can be integrated into or on the electronic component.

[0032] Examples and embodiments of the present invention will be illustrated below.

[0033] With reference to Figures 1 to 6 of the attached drawing: Figure 1 shows a flow diagram of a closed temperature control system for an electronic component inv,

[0034] Figure 2 shows a possible breakdown of the individual components required for this, Figure 3 shows a load profile P in kilowatts as a function of time t in seconds,

[0035] Figure 4 shows a temperature profile T in °C as a function of time t in seconds.

[0036] Figure 5 shows the relative lifetime extension Inc (%) as a function of the current power output relative to the nominal power output in percent Ref (%) and

[0037] Figure 6 shows the mean efficiency Eff (%) as a function of the current power in relation to the nominal power Ref in percent (%).

[0038] Figure 1 shows the closed-loop control system for temperature adjustment of an electronic component inv. It consists of the following components: first, the AI-based load forecast Al-Pf, a load forecast model. This component is, in particular, a software component or contained within a software component. A series of actual values ​​P / T are input parameters into the AI-based load forecast Al-Pf. It then provides a power forecast Pf for a specific period. Depending on the application or the function of the electronic component inv, a typical useful forecast period is, for example, between 5 and 60 minutes. The forecast period should be sufficiently long to identify "short" troughs in the load profile P(t), see Figure 3, for which it is advantageous to apply temperature control T-con for compensation.

[0039] The actual values ​​required are the actual load P of the electronic component, for example, an inverter inv, and a temperature measurement T, which temperature measurement T also serves as the input value for the temperature adaptation control T-con. Since the typical load cycles P(t) of an inverter inv depend on time t and date d, these are also fed in as additional parameters in a preferred embodiment. Furthermore, other application-specific load parameters can be added, such as weather forecasts and sun position, whether the current day is a working day, planned load schedules of consumers, or power sources in an island grid. In general, a range of necessary and sufficient data should be provided to generate the most accurate possible load forecast Pf.

[0040] In addition to current values, historical data from a specific past period is also required so that the AI ​​can generate accurate forecasts. This data is collected and processed internally using suitable network architectures such as LSTM (Long-Short-Term Memory).

[0041] Another component of the closed-loop control system for temperature adjustment of an electronic component (inv) is the training (tr) of the load prediction model (Al-Pf). For example, the load prediction model (Al-Pf) is continuously trained (tr) during operation via online learning. In this process, the actual load (P) is compared with the previously generated load prediction (Pf). The prediction error is preferably minimized by using a supervised learning method. The time required for weight updates due to online learning is not critical in this application, as predictions (Pf) do not necessarily need to be provided at fixed time intervals. Therefore, relatively long update intervals of up to one minute are possible. Furthermore, additional parameters (cf), such as a user configuration or the electricity price, can be included. The temperature simulation component (T-sim) is also integrated.The load forecast profile Pf is fed into this temperature simulation T-sim. Based on the actual temperature T, another input parameter, and the load forecast profile Pf, the temperature simulation T-sim generates a temperature-load profile. Typically, the temperature measurement T represents the heatsink and / or power module temperature of the electronic component inv. Furthermore, all individual die temperatures of the IGBTs and diodes can be calculated using a thermal model. Such models can be derived, for example, from a 3D temperature simulation that considers all device couplings while taking model reduction into account. Based on a load forecast Pf, an individual temperature profile T(t) could therefore be generated for each die in a power module inv.For applications other than power electronics devices, this component inv can be replaced by a suitable physical or data-based simulation model to calculate important temperatures T.

[0042] Another component of the closed-loop control system for temperature adjustment of an electronic component inv is the temperature control T-con. Considering the temperature forecast profile Tf, the goal of the temperature control T-con is not to maintain a constant energy level, as this can be inefficient. Instead, the temperature control T-con focuses on reducing large temperature fluctuations, as shown in the temperature profile T(t) in Figure 4. This is achieved primarily through efficient management of the heating process, particularly by intervening in the cooling process. One approach, for example, involves optimizing the cooling slope, especially in response to anticipated high-load conditions. This optimization varies depending on whether the system uses active or passive cooling.With active cooling, adjustments can be made by controlling the liquid pump flow rate or fan speed. In the case of passive cooling, methods such as reactive power control, energy dissipation, or the application of on-demand load can be used to fine-tune the cooling slope. The decision-making process, which considers these different factors and aims to effectively reduce the delta temperature, is complex. This is where AI (artificial intelligence) plays a crucial role. AI algorithms help solve this two-criteria problem: minimizing temperature fluctuations to extend the device's lifetime while simultaneously ensuring energy efficiency. This is achieved by determining the most intelligent cooling strategy in the context of the current load profile P(t) and the available cooling options.The mean efficiency Eff (%) and relative lifetime increase Inc (%) are plotted in the diagrams in Figures 5 and 6 as a function of the current power output relative to the nominal power Ref (%). The power output here refers to the additional power used to reduce cooling behavior and not to the power output of a device. If the temperature control T-con is based on a lifetime model for the device inv that converts a temperature cycle into a monetary value, for example, taking into account the cost of replacing the device, this problem reduces to a one-dimensional convex continuous optimization problem that can be solved conventionally, for example, using line-search methods.

[0043] Typically, the T-con temperature control requires the temperature forecast profile Tf as input. Alternatively, the load forecast profile Pf or both the load and temperature forecast profiles Tf+Pf can be used.

[0044] Finally, the inv-con controller of the electronic component inv, specifically an inverter controller, is another component of the closed-loop control system for temperature adaptation. This inv-con controller is configured to receive direct setpoints from the temperature controller T-con to manage thermal and electrical performance parameters. These setpoints, aimed at optimizing the temperature profile T(t), can include adjustments to reactive power, variations in switching frequency, and modulation of switching speed. Instead of processing the corresponding data internally, the inv-con controller acts according to these specific instructions. In particular, this modifies the conventional, efficiency-optimized operations of the inv-con controller.

[0045] To further improve the interaction between the temperature controller T-con and the component controller inv-con, the component inv communicates its current operating parameters to the temperature controller T-con. This communication enables the temperature controller T-con to consider these operating parameters and requirements when determining the optimal temperature profile and the corresponding setpoints.

[0046] The component controller inv-con can therefore adapt its operations based on the temperature controller T-con, for example by generating additional thermal losses through frequency control or by using reactive power control. The component controller inv-con returns the actual temperature values ​​T to the temperature controller T-con and to the temperature simulation model T-sim. Furthermore, it communicates the current operating states P to the control module for training and updating the forecast Pf.

[0047] The Kl model Al-Pf is continuously updated with current operating data T and P. It regularly updates the load profile predictions Pf, with the update interval being significantly shorter than the prediction horizon. This means that approximately one update per minute for a 15-minute time horizon would be a good solution, taking into account computational effort, thermal management accuracy, and the performance of this closed-loop control system. Whenever a new update is available, it is processed by the temperature simulation T-sim and the temperature control T-con to provide updated recommendations to the component control inv-con. A particular advantage of the proposed solution lies in its iterative corrective approach, which synergizes real-time data analysis, cloud-based digital twin technology, and machine learning algorithms. This ensures that an accurately predicted temperature level is maintained.The derivation of optimal operating conditions is facilitated, aiming to improve service life and optimize efficiency by considering various parameters such as investment costs, energy costs, and CO2 footprint to obtain the most sustainable and economically viable solution. See Figures 3 to 6 for further information.

[0048] Figure 3 shows a predicted power output P(kW), and Figure 4 shows the resulting predicted temperature T(°C), specifically a heat sink or component temperature, as a function of time t. Figure 5 shows the increase in lifetime Inc (%) compared to the loss in efficiency Eff (%) in Figure 6 as a function of the minimum power reference value Ref in %. The proposed solution integrates condition monitoring, thereby enabling demand-driven maintenance, which leads to further advantages:

[0049] The use of real-time data (T, P) in combination with the predictive capabilities of machine learning algorithms (Al-Pf) significantly improves the quality of system performance forecasts (Pf). This enables proactive adjustments during operation to identify and resolve potential problems early on. By continuously analyzing operating conditions and adapting strategies accordingly, the proposed solution maximizes the lifetime (Inc) of the controlled electronic devices (inv). This is achieved through optimal management of operating loads (inv-con) and temperatures (T-con). This minimizes factors that frequently lead to premature wear and failures. Furthermore, it ensures more efficient maintenance. The possibility of on-demand maintenance, supported by continuous condition monitoring, ensures that service and repairs are performed when needed, rather than according to a predetermined schedule.This approach not only saves resources but also prevents unnecessary downtime of the electronic device and improves the overall efficiency of the system.

[0050] Operational effectiveness is also improved. Operating parameters T and P are dynamically adjusted to changing conditions, leading to more efficient use of energy and resources. This is particularly advantageous in scenarios where power demand is variable and unpredictable. Furthermore, enhanced analytical capabilities are provided: The cloud-based digital twin approach offers a sophisticated analytics platform that enables a deeper and more comprehensive understanding of operational dynamics and the impact of various factors on system performance P. Based on a holistic, data-driven approach, the proposed solution not only ensures optimal efficiency Eff and extended service life Inc, but also delivers significant improvements in predictability, maintenance efficiency, and overall operational effectiveness.This contributes to sustainability and establishes a data flow between the manufacturing area, the edge, and optionally the cloud for intelligent, software-based automation and IT / OT integration. In summary, the proposed computer-implemented method has the following aspects: the load control transmits current temperature measurement data of the electronic component to the temperature regulation and temperature simulation systems; current load measurement data and temperature measurement data of the electronic component are also transmitted to AI-based load prediction and AI training. Extending the service life is prioritized over immediate efficiency maximization, as service life has become a critical design criterion in all electronics. This is achieved through temperature regulation of the electronic component via load control and optional active thermal management.

[0051] Reference symbol list t time, point in time d date, further application-specific parameters tr KL training

[0052] Al-Pf AI- or Kl-based load forecasting

[0053] Pf Load Forecast

[0054] T-sim temperature simulation

[0055] Temperature forecast (Tf) and other parameters, for example user configuration, electricity price

[0056] T-con temperature regulation, temperature control; inv-con load control of the electronic component; inv electronic component, for example inverter

[0057] SW software component, especially on an edge device

[0058] HW hardware component, especially electronic component with integrated load and

[0059] Temperature control

[0060] P, T Power and temperature measurement data t(s) Time in seconds

[0061] P(kW) Power in kilowatts

[0062] T (°C) Temperature in degrees Celsius

[0063] Ref (%) current output in relation to nominal output in percent

[0064] Eff (%) Average efficiency

[0065] Inc (%) Increase of Lifetime, relative lifespan extension

Claims

Patent claims 1. A computer-implemented method comprising temperature control (T-con) of an electronic component (inv), in particular a power electronic component, wherein a load forecast (Pf) for the electronic component (inv) is incorporated into a temperature simulation (T-sim) for the temperature control (T-con), by means of which a temperature prediction (Tf) over a certain period of time is determined for the electronic component (inv), and wherein load measurement data (P) and temperature measurement data (T) of the electronic component (inv) are also incorporated into the temperature simulation (T-sim), wherein the load forecast (Pf) is created using AI (Al-Pf), and based on the load forecast (Pf) and temperature prediction (Tf), temperature control (T-con) of the electronic component (inv) is carried out via the load control (inv-con).

2. Computer-implemented method according to claim 1, wherein the load control (inv-con), based on setpoints received from the temperature control (T-con), controls thermal as well as electrical power parameters of the electronic component (inv).

3. Computer-implemented method according to claim 2, wherein the load control (inv-con) particularly comprises an adjustment of reactive power, a variation of a switching frequency or a modulation of a switching speed.

4. Computer-implemented method according to one of the preceding claims, wherein the electronic component (inv) communicates values, in particular operating limits, to the temperature control (T-con).

5. Computer-implemented method according to one of the preceding claims, wherein the temperature measurement data (T) which are included in the temperature simulation (T-sim) comprise at least the temperature of a component of the electronic component (inv), in particular the temperature of a cooling component.

6. Computer-implemented method according to claim 5, wherein temperature values ​​(T) of further individual components of the electronic component (inv), in particular individual IGBTs and / or diodes, are calculated using a thermal model.

7. Computer-implemented method according to one of claims 5 or 6, wherein an individual temperature profile is generated based on the load prediction (Pf) for several components within the electronic component (inv).

8. Computer-implemented method according to one of the preceding claims, wherein the temperature regulation (T-con) of the electronic component (inv) is carried out by means of active thermal management.

9. Computer-implemented method according to claim 8, wherein the active thermal management comprises a cooling process which in turn comprises controlling a liquid pump flow, controlling a fan speed, controlling reactive power, controlling energy output and / or applying a distributable load for fine-tuning the cooling capability.

10. Computer-implemented method according to claim 8 or 9, wherein a two-criteria problem, in particular from temperature fluctuation minimization and energy efficiency or energy availability, is solved, in particular by means of a KL algorithm, and the active thermal management is adapted to the cooling capability of the cooling system used in the electronic component (inv).

11. Computer-implemented method according to one of the preceding claims, wherein a data input (P, T, tr) is provided to the AI-based load forecast (Al-Pf), in particular a data input comprising a data history.

12. Computer-implemented method according to one of the preceding claims, wherein measurement data (T, P) of the electronic component (inv) are input into the KL-based load prediction (Al-Pf).

13. Computer-implemented method according to one of the preceding claims, characterized in that the load prediction class (Al-Pf) is continuously trained (tr) during the execution of the method.

14. Computer-implemented method according to claim 13, wherein an input of load measurement data (P) and / or temperature measurement data (T) of the electronic component (inv) is made into the KL training (tr).

15. Arrangement for carrying out a computer-implemented method according to one of claims 1 to 14, comprising a hardware component (HW), in particular comprising an electronic component (inv) with load control (inv-con) and temperature regulation (T-con), and a software component (SW), in particular comprising temperature simulation (T-sim) and AI-based load prediction (Al-Pf).

Images