Predictive model resulting from machine learning for predicting the temperature of an item of equipment

Abstract

The invention relates to a method for predicting a temperature of at least one first point of interest of an item of equipment, the method comprising the following steps: - obtaining (204) a set of current input data comprising at least one current input datum in relation to the item of equipment, the current input datum being representative of a control parameter of the item of equipment; - applying (205) a predictive model to the set of current input data in relation to the item of equipment and to at least one current temperature of the item of equipment, said predictive model resulting from machine learning applied to a set of training data; - following the application of the predictive model, predicting (206) a future temperature of the first point of interest of the item of equipment.

Description

Machine learning-based predictive model for equipment temperature prediction

[0001] The present invention relates to the field of temperature prediction in equipment, in particular for monitoring the performance of such equipment. The invention applies in particular, but not exclusively, to automotive equipment such as lighting devices.

[0002] Monitoring the temperature in equipment, the operation of which leads to heat dissipation, helps ensure the proper functioning of the equipment, and to verify that the temperature of the equipment remains within a predetermined operating range. A piece of equipment may also include one or more points of interest whose temperature it is preferable to be able to monitor.

[0003] Such temperature monitoring can be achieved by using a temperature sensor dedicated to the point of interest. Such a sensor can be an NTC thermistor, for "Negative Temperature Coefficient" in English. However, such a temperature sensor is expensive, with parts made of gold, and increases the size, especially when the points of interest are on small electronic components, as is the case on printed circuit boards, or PCBs, for "Printed Circuit Board" in English.

[0004] Patent application US4656829A provides for replacing the measurement of parameters, such as temperature, with a prediction based on parameters other than temperature, to which a mathematical model is applied. However, the development of such a model is specific to the equipment considered, and is very complex and expensive to develop, especially when the relationship with temperature is not linear. A simpler model would lead to a temperature prediction defect, which would cause over time too large a deviation from the actual temperature value.

[0005] There is therefore a need to accurately predict the temperature of at least one point of interest in a piece of equipment, while reducing the number of sensors in the equipment as well as the complexity and costs associated with the prediction model.

[0006] The present invention improves the situation.

[0007] To this end, a first aspect of the invention relates to a method for predicting a temperature of at least one first point of interest of a piece of equipment, the method comprising the following steps: obtaining a set of current input data comprising at least one current input data of the equipment, the current input data being representative of a control parameter of the equipment; applying a predictive model to the set of current input data of the equipment and to at least one current temperature of the equipment, the predictive model being derived from machine learning applied to a set of training data; following the application of the predictive model, predicting a future temperature of the first point of interest of the equipment.

[0008] Using a machine-learning predictive model to predict the future temperature at a first point of interest allows for the input of a current temperature of the equipment that is different from a current temperature measured at the first point of interest. This makes it possible to access the temperature of a point of interest without necessarily placing a temperature sensor there. Moreover, such a predictive model can be developed for different equipment, by modifying the training data set on which the predictive model is based. It is therefore much less expensive and complex to develop than a mathematical model dedicated to one piece of equipment.

[0009] According to one embodiment, the current temperature of the equipment to which the predictive model is applied may be a current temperature of the first point of interest predicted by the predictive model during a previous iteration.

[0010] Thus, the predictive model makes it possible to completely avoid temperature sensors in the equipment, the current temperature of a point of interest being the temperature predicted in the previous iteration. The initialization of temperatures can be predetermined or based on an external ambient temperature sensor. This makes it possible to avoid the need for expensive and sometimes complex temperature sensors to install.

[0011] Additionally, the predictive model can be applied to the current input data set, to the current temperature of the first point of interest estimated by the predictive model during the previous iteration, and to a current temperature of a second point of interest of the equipment, different from the first point of interest.

[0012] Thus, the model can take as input a current state vector, each component of which is a current temperature of a point of interest of the equipment. It is thus made possible to model the equipment by several points of interest, in particular when the equipment includes several modules or components controlled separately, and which can have temperatures with different evolutions.

[0013] Additionally, the current temperature of the second point of interest of the equipment can be measured by a temperature sensor dedicated to the second point of interest of the equipment.

[0014] Measuring a temperature at a second point of interest is a feedback data, improving the accuracy associated with the prediction, especially after several iterations of the process, which delays the divergence of the model from the actual temperature values. In addition, this gain in accuracy is possible without having to provide a sensor at the first point of interest.

[0015] Alternatively, the current temperature of the second equipment point of interest may be predicted by the predictive model in a previous iteration.

[0016] This makes it possible to predict the temperature at multiple points of interest, without requiring the placement of sensors at these points of interest, which significantly reduces the number of temperature sensors, and therefore the costs associated with the equipment.

[0017] According to one embodiment, following application of the predictive model, the method comprises predicting a first future temperature of the first point of interest of the equipment and a second future temperature of the second point of interest.

[0018] Thus, the predictive model is able to predict several future temperatures for several points of interest, which is advantageous in equipment that presents spatial variations in temperature, particularly for equipment comprising several functions or several modules. A single predictive model is thus used to predict the temperature at several points of interest, which makes it possible to share the costs associated with training the predictive model.

[0019] According to one embodiment, the equipment may be a lighting device of a motor vehicle, and the set of current input data may comprise at least one driving current value of a module of the lighting device.

[0020] The invention applies advantageously to the context of a lighting device for a motor vehicle, for which cost and space constraints are significant.

[0021] Additionally, the first point of interest and the second point of interest may be located in first and second modules of the lighting device, respectively, and the set of current input data may include a first drive current value of the first module and a second drive current value of the second module.

[0022] This makes it possible to predict the temperature in modular equipment, performing several functions. The predictive model thus makes it possible to dedicate a point of interest to each of the equipment's modules.

[0023] Alternatively, the first point of interest and the second point of interest may be located in the same module, and may be associated respectively with a first light source of the module and a second light source of the module, the first and second light sources being associated with distinct lighting or signaling functions, and the set of current input data may comprise a first driving current value of the first light source and a second driving current value of the second light source.

[0024] Thus, this embodiment makes it possible to accurately measure the temperature in several points of interest of the same module, which is particularly advantageous when the same module comprises several sets of sources dedicated to distinct functions.

[0025] According to one embodiment, the method may comprise a prior step of storing the predictive model in a control module of the equipment, the control module being able to access the set of current input data of the equipment.

[0026] Thus, the same entity is responsible for controlling the equipment and predicting the temperature at the first point of interest, which simplifies the implementation of the invention, and allows the control module to adapt the control of the equipment according to the predicted future temperature.

[0027] According to one embodiment, the predictive model can be obtained by a prior supervised learning step based on training data, the training data comprising a training set of input data of the equipment corresponding respectively to respective temperature measurements of the point of interest of the equipment.

[0028] The training data is thus specific to the equipment, which improves the accuracy associated with the prediction.

[0029] According to one embodiment, the predictive model can further be applied to a current ambient temperature measurement from an ambient temperature sensor.

[0030] In this way, a return or feedback value is taken into account, which improves the accuracy associated with the prediction, without having to install a temperature sensor in the equipment.

[0031] A second aspect of the invention relates to a computer program comprising instructions for implementing the method according to the first aspect of the invention, when these instructions are executed by a processor.

[0032] A third aspect of the invention relates to a control module of an equipment, comprising

[0033] a memory storing a predictive model resulting from machine learning applied to a training data set

[0034] a processor configured to obtain a set of current input data comprising at least one current input data of the equipment, the current input data being representative of a control parameter of the equipment; apply the predictive model to the set of current input data of the equipment and to at least one current temperature of the equipment; following the application of the predictive model, prediction of a future temperature of the first point of interest of the equipment.

[0035] Other characteristics and advantages of the invention will appear on examining the detailed description below, and the appended drawings in which:

[0036] illustrates a system for predicting the temperature of a point of interest of equipment according to an embodiment of the invention;

[0037] is a diagram illustrating the steps of a method for predicting the temperature of a point of interest of equipment according to an embodiment of the invention;

[0038] illustrates a structure of a control module of equipment according to an embodiment of the invention.

[0039] The description focuses on the characteristics that distinguish the method or the set from those known in the state of the art.

[0040] Illustrates a system for predicting a temperature of a point of interest of a piece of equipment 101, according to an embodiment of the invention.

[0041] The equipment illustrated in is a lighting device for a motor vehicle. Such an example is given for illustrative purposes only and the invention applies, however, to any equipment capable of being controlled by a set of input data comprising at least one input data item. Input data is understood to mean data representative of a control parameter of the equipment. In the case of the lighting device 101, an input data item may be the value of the current delivered to a lighting module of the lighting device 101. In particular, the invention may advantageously be applied to equipment whose operation generates heat, and thus induces local temperature variations. The invention may thus be applied to equipment other than the lighting device of a motor vehicle, such as the vehicle engine or the heating / air conditioning system.Alternatively, the invention may be applied to equipment not related to a motor vehicle, no restriction being attached to such equipment.

[0042] In the example of the, the lighting device 101 comprises two lighting modules, namely a first module 102.1 and a second lighting module 102.2.

[0043] The first module 102.1 may be a module comprising LED type light sources for example, the light sources being separated into a first set of light sources 103.1 dedicated to a dipped beam function, also called “Low Beam” in English, or LB, and a second set of light sources 103.2 dedicated to a main beam function, also called “High Beam” in English, or HB.

[0044] Such a first 102.1 module can be called Multi-LEDs.

[0045] The second module 102.2 may be a module having a pixelated source dedicated to one or more functions other than the LB and HB functions described above.

[0046] It will thus be understood that the lighting device 101 is a particular type of lighting device, but that the invention applies to lighting devices of different types, with varied technologies and sources of different natures than those described with reference to the. The lighting device may for example comprise high definition monolithic type sources, comprising a large number of LEDs, for example greater than 100 LEDs, or a DMD type source, for “Digital Micro-mirror Device” in English.

[0047] Each of the modules, or even subsets of sources of a module for the first module 102.1, can be controlled by a control module 110, or ECU for Electronic Control Unit, in English. Such an ECU 110 can be dedicated to the lighting device 101 or can also perform other functions in the motor vehicle, in which case the vehicle can comprise a single centralized ECU, controlling all of the equipment of the motor vehicle.

[0048] The ECU 110 is able to control the modules, or even individually the sources of each module, via respective input currents. Such currents are the input data of the lighting device 101, since they are the parameters used for controlling the lighting device 101. Other data influence its operation, such as the ambient temperature, the temperature of points of interest, vibrations, but are not input data because they are not directly controlled by the ECU 110.

[0049] There are no restrictions on what is meant by a point of interest. It can be a point near or on a component of the equipment whose temperature monitoring is particularly relevant, because the component is likely to heat up or because its operation is essential to the equipment.

[0050] For example, in the example of the, the following points of interest can be considered: a first point of interest 104.1 inside the first module 102.1, in particular near the first set of sources 103.1 dedicated to the LB function; a second point of interest 104.2 also inside the first module 102.1, near the second set of sources 103.2 dedicated to the HB function; a third point of interest 104.3 inside or near the second module 102.2.

[0051] Other points of interest may be provided. For example, the second module 102.2 may comprise several points of interest and the first module 102.1 may comprise only one, or more than three. Furthermore, another point of interest may correspond to the control module 110, whose temperature may be relevant to predict or measure.

[0052] The invention provides for estimating or measuring the temperature in at least one point of interest, by using a reduced number of temperature sensors, in particular less than the number of points of interest, or even without using the slightest temperature sensor in the lighting device 101.

[0053] The invention also applies to embodiments with a single point of interest.

[0054] The system 100 according to the invention may also comprise an ambient temperature sensor 105 capable of measuring an ambient temperature, that is to say a temperature of a point located at a distance from the first and second modules 102.1 and 102.2, and from the control module 110, for example outside the lighting device 105. No restriction is attached to the exact location of the ambient temperature sensor, which is moreover optional according to the invention.

[0055] This is a diagram illustrating the steps of a method according to one embodiment of the invention.

[0056] The method for predicting the temperature of a point of interest according to the invention applies to any equipment. However, it is described below in the context, given for illustrative purposes, of the lighting device 101 of the.

[0057] Steps 200 to 202 are preliminary steps for constructing a predictive model according to the invention.

[0058] The predictive model aims to predict the future temperature of at least one point of interest, based on a current temperature of the lighting device 201, and based on the current input data of the lighting device 201.

[0059] As specified above, the input data are the control currents of the different functions, namely: a first current controlling the first set of sources 103.1, for the LB function, the first current being noted u1(t)=LB_Current; a second current controlling the second set of sources 103.2, for the HB function, the second current being noted u2(t)=HB_Current; a third current controlling the second module 102.2 with pixelated source, for another function, the third current being noted u3(t)=Pixel_Current.

[0060] The input data set is denoted u(t), and can be considered as a vector comprising the three components u1(t), u2(t) and u3(t).

[0061] According to the invention, the set of input data comprises at least one input data. For example, a lighting device comprising only the second module 102.2 would have a single input data u3(t).

[0062] The input data are controlled, or at least accessible, by the control module 110 according to the invention.

[0063] A set of output data, denoted y(t) of the lighting device 101 are the luminous fluxes emitted for each function, namely: a first luminous flux y1(t) for the LB function; a second luminous flux y2(t) for the HB function; and a third luminous flux y3(t) for the other function of the second module 102.2.

[0064] More generally, the output data of a piece of equipment are representative of the results obtained and sought by the implementation of the equipment's functions. For example, the equipment also generates heat, but this is not a desired result, so the generated heat is not part of the output data.

[0065] In the remainder of the description, the temperatures of the set of points of interest of the equipment, which includes at least one point of interest, and optionally the ambient temperature, are referred to as "states".

[0066] In the case of the lighting device 110, a state vector x(t) may comprise the following components: a first state x1(t)=LB_Temp representing the temperature of the first point of interest 104.1; a second state x2(t)=HB_Temp representing the temperature of the second point of interest 104.2; a third state x3(t)=Pixel_Temp representing the temperature of the third point of interest 104.3; a fourth state x4(t)=Ambient_Temp representing the temperature measured by the ambient temperature sensor 105, the presence of which is optional as described above.

[0067] The predictive model aims to determine a function f() such that: x(t+1)=f(x(t), u(t)).

[0068] Developing a mathematical model is both complex and must be repeated for each piece of equipment. Thus, a change in the lighting system requires the development of a new mathematical model. Moreover, developing such a mathematical model is time-consuming and costly.

[0069] The present invention proposes to use machine learning to develop the f() model.

[0070] In a step 200, a training data set is formed. The training data set is a set of real or simulated data obtained for the lighting device 101. The training data set comprises both training input data u(t), namely the current values, but also the states x(t) corresponding to each input data u(t).

[0071] The training data set is constituted so as to be representative of most environmental condition situations to which the lighting device is exposed. For example, the training data can be constituted to vary the following parameters: the ambient temperature. For example, three ambient temperature levels can be set at 25°, 42.5° and 70°C; the activation or not of a ventilation system of the lighting device 101; the activation of a combination of lighting functions, HB, LB or other; the current levels, or input data, applied; the control mode, PWM type or not, and if PWM, the values ​​of the duty cycles applied.

[0072] No restrictions on the parameters used to vary the training data.

[0073] Following the obtaining of training data in step 200, a model is trained by machine learning in a step 201, based on the training data. No restriction is attached to the type of model trained, which can be a RidgeCV type model or a neural network, the number of neural layers of which depends on the desired accuracy and constraints related to complexity.

[0074] Any other machine learning model may be used according to the invention.

[0075] The principles of machine learning are well known and are not detailed further in this application.

[0076] A predictive model is thus obtained at a step 202, the model being capable of predicting future states x(t+1) on the basis of current states x(t) and current input data u(t).

[0077] The following steps of the method according to the invention are implemented in the control module 110 introduced with reference to the.

[0078] In a step 203, the predictive model is stored in a memory of the control module 110.

[0079] In a step 204, current input data u(t) are obtained by the control module 110. As indicated above, the current input data u(t) can in particular be controlled by the control module 110, which thus has direct access to it.

[0080] At a step 205, the predictive model is applied to the current input data u(t) and to the temperatures of the current states x(t).

[0081] Several embodiments are provided for the implementation of step 205, depending in particular on the number of sensors allocated to the points of interest, the number of sensors possibly being zero according to certain embodiments of the invention.

[0082] In a so-called open-loop embodiment, none of the three points of interest 104.1, 104.2 and 104.3 is equipped with a dedicated temperature sensor. Consequently, no temperature sensor is provided in the equipment 101, only a temperature value can be received from the ambient temperature sensor 105 as feedback.

[0083] The open-loop implementation thus makes it possible to reduce the number of sensors in the lighting device as much as possible, while still having access to the temperature of several points of interest, thanks to the predictive model obtained by machine learning.

[0084] In this open-loop embodiment, the current state vector, comprising the current temperatures for points of interest 104.1, 104.2 and 104.3, corresponds to the temperature values ​​estimated during the previous iteration of the method, when t was equal to t-1.

[0085] In this case, the state vector x(t) subject to the predictive model is the following vector:x1(t) = f1(x(t-1), u(t-1));x2(t) = f2(x(t-1), u(t-1));x3(t) = f3(x(t-1), u(t-1));

[0086] f1 being the output of the predictive model for estimating the temperature for the first point of interest 104.1, f2 being the output of the predictive model for estimating the temperature for the second point of interest 104.2, and f3 being the output of the predictive model for estimating the temperature for the third point of interest 104.3.

[0087] The fourth component of the state vector x4(t) can be taken from the ambient temperature sensor 105, and x4(t)=Ambient_Temp.

[0088] The current state vector x(t) can thus be used to predict the future state vector x(t+1) by the model x(t+1) = f(x(t), u(t)). Note that alternatively, the predictive model can take into account only some of the components of the state vector, this subset of components being noted x'(t).

[0089] In this case, the predictive model applies as follows: x(t+1) = f(x'(t), u(t)).

[0090] During the first iteration of the method according to the open-loop embodiment of the invention, the temperatures of the different points of interest can be initialized from predefined temperatures, or from the ambient temperature measured by the sensor 105. Indeed, the method can be initiated when the equipment is started, or when a vehicle, such as a motor vehicle comprising the equipment, is started. At initialization, the temperature of the equipment is therefore close to the ambient temperature.

[0091] Three closed-loop embodiments are described below, in which at least one feedback value is obtained from at least one temperature sensor located in the lighting device 101.

[0092] In a first closed-loop embodiment, three feedback values ​​are received from three temperature sensors respectively dedicated to the three points of interest 104.1, 104.2 and 104.3.

[0093] Thus, the state vector x(t) used as input to the predictive model may include the following components, or a subset of the following components when the model takes x'(t) as input: x1(t) = LB_Temp, current temperature measured by a first sensor associated with the first point of interest 104.1; x2(t) = HB_Temp, current temperature measured by a second sensor associated with the second point of interest 104.2; x3(t) = Pixel_Temp, current temperature measured by a third sensor associated with the third point of interest 104.3; x4(t) = Ambient_Temp.

[0094] The first closed-loop embodiment does not allow to reduce the number of sensors, which is equal to the number of points of interest, but allows to evaluate the predictions of the predictive model, and to compare the performance of the other embodiments with the first embodiment, which is an ideal case.

[0095] In a second closed-loop embodiment, one of the points of interest does not include a temperature sensor, and its future temperature at t+1 is predicted from the predictive model, current input data, feedback data from sensors placed on the other points of interest, and its current temperature at t predicted at the previous iteration. For example, the point of interest that does not include a dedicated temperature sensor is the first point of interest 104.1, and the current state vector x(t) taken as input to the predictive model may include the following components, or a subset of the following components when x'(t) is taken as input to the predictive model: x1(t) = f1(x(t-1), u(t-1)), current temperature of the first point of interest 104.1 predicted by the model during the previous iteration; x2(t) = HB_Temp, current temperature measured by a first sensor associated with the second point of interest 104.2 ;x3(t) = Pixel_Temp, current temperature measured by a second sensor associated with the third point of interest 104.3 ;x4(t) = Ambient_Temp.

[0096] Thus, according to the second closed-loop embodiment, the number of temperature sensors is reduced compared to the first closed-loop embodiment. The number of sensors in the equipment is thus less than the number of points of interest whose respective temperatures are predicted.

[0097] In a third closed-loop embodiment, at least two of the points of interest do not include a temperature sensor, and their future temperatures at t+1 are predicted from the predictive model, current input data, feedback data from sensors located on the other points of interest and their current temperatures at t estimated by the predictive model at the previous iteration. For example, the points of interest that do not include a dedicated temperature sensor are the first point of interest 104.1 and the second point of interest 104.2, and the current state vector x(t) taken as input to the predictive model, may comprise the following components, or a subset of the following components when x'(t) is taken as input to the predictive model: x1(t) = f1(x(t-1), u(t-1)), current temperature of the first point of interest 104.1 predicted by the model during the previous iteration;x2(t) = f2(x(t-1), u(t-1)), current temperature of the second point of interest 104.2 predicted by the model during the previous iteration;x3(t) = Pixel_Temp, current temperature measured by a first sensor associated with the third point of interest 104.3;x4(t) = Ambient_Temp.

[0098] The third closed-loop embodiment thus makes it possible to predict the future temperature of three points of interest from a single temperature sensor. Moreover, the temperature sensor is placed in the second module 102.2 separate from the first module 102.1 in which the first and second points of interest not having a temperature sensor are present.

[0099] It will be understood that other closed-loop embodiments can be provided, in particular when the number of points of interest of the equipment is strictly greater than 3.

[0100] Following step 205, the vector of future states x(t+1) is obtained, the components of which are the future temperatures at the different points of interest. In particular, the vector of future states may comprise a future temperature associated with a point of interest to which no sensor is allocated. The invention thus advantageously makes it possible to reduce the number of sensors, or even to dispense with any sensor in the equipment.

[0101] At a step 207, the method can be iterated and the future time t+1 becomes the current time t. The method then returns to step 204 to receive new current input data u(t).

[0102] In practice, comparable results are obtained for the open-loop embodiment, and for the three closed-loop embodiments. In particular, each embodiment makes it possible to maintain a low deviation from the actual temperature values, over a long prediction period, greater than several hours, which makes the invention compatible with most equipment whose operating times are less than one day.

[0103] This presents a structure of the control module 110 of the equipment 101, according to an embodiment of the invention.

[0104] The control module 110 comprises a processor 301 configured to communicate unidirectionally or bidirectionally, via one or more buses or via a wired connection, with a memory 302 such as a “Random Access Memory” type memory, RAM, or a “Read Only Memory” type memory, ROM, or any other type of memory (Flash, EEPROM, etc.). Alternatively, the memory 302 comprises several memories of the aforementioned types. Preferably, the memory 302 is a non-volatile memory.

[0105] The memory 302 is capable of storing, permanently or temporarily, all of the data generated following the implementation of the method described above. In particular, the memory 402 can store the predictive model during the step 203 described above. The memory 302 can also store the temperatures predicted during the previous iterations to make a prediction of a future temperature during a current iteration.

[0106] The processor 301 is capable of executing instructions, stored in the memory 302, for implementing steps 203 to 207 of the method illustrated with reference to the. Alternatively, the processor 301 can be replaced by a microcontroller designed and configured to carry out steps 203 to 207 of the method according to the invention.

[0107] The control module 110 may comprise an input interface 303 capable of obtaining the input data from the equipment 110, in particular the currents of the different modules and set of light sources in the case of the lighting device 101. However, such an input interface 303 is optional since the control module 110 can directly access the input data when it is in charge of controlling the equipment 101.

[0108] No restriction is attached to the first input interface 303 which can be a wired interface for example, or alternatively wireless.

[0109] The control module 110 may further comprise a second output interface 304 capable of communicating the predicted temperatures for the point(s) of interest to an entity external to the control module 110. Such an interface 304 is optional. Alternatively, the predicted temperatures are used by the control module 110 to adapt the input data used for controlling the equipment.

[0110] The present invention is not limited to the embodiments described above as examples; it extends to other variants.

Claims

Method for predicting a temperature of at least one first point of interest (104.1 -104.3) of a piece of equipment (101), the method comprising the following steps:- obtaining (204) a set of current input data comprising at least one current input data of the equipment, the current input data being representative of a control parameter of the equipment;- application (205) of a predictive model to the set of current input data of the equipment and to at least one current temperature of the equipment, said predictive model being derived from machine learning applied to a training data set;- following the application of the predictive model, prediction (206) of a future temperature of the first point of interest of the equipment. Method according to claim 1, wherein the current temperature of the equipment (101) to which the predictive model is applied, is a current temperature of the first point of interest (104.1-104.3) predicted by the predictive model during a previous iteration. Method according to claim 2, wherein the predictive model is applied to the current input data set, to the current temperature of the first point of interest (104.1-104.3) predicted by the predictive model during the previous iteration, and to a current temperature of a second point of interest (104.1-104.3) of the equipment (101), different from the first point of interest. Method according to claim 3, wherein the current temperature of the second point of interest (104.1-104.3) of the equipment (101) is measured by a temperature sensor dedicated to the second point of interest of the equipment. Method according to claim 3, wherein the current temperature of the second point of interest (104.1-104.3) of the equipment (101) is predicted by the predictive model during a previous iteration. A method according to any one of claims 3 to 5, wherein, following the application of the predictive model, the method comprises the prediction (206) of a first future temperature of the first point of interest (104.1-104.3) of the equipment and of a second future temperature of the second point of interest. A method according to any one of the claims, wherein the equipment (101) is a lighting device for a motor vehicle, and wherein the current input data set includes at least one drive current value for a module (102.1; 102.2) of the lighting device. A method according to claim 7 and any one of claims 3 to 6, wherein the first point of interest (104.1; 104.2) and the second point of interest (104.3) are located respectively in first and second modules of the lighting device, and wherein the set of current input data includes a first drive current value of the first module (102.1) and a second drive current value of the second module (102.2). A method according to claim 7 and any one of claims 3 to 6, wherein the first point of interest (104.1) and the second point of interest (104.2) are located in the same module (102.1), and are associated respectively with a first light source of the module and a second light source of the module, the first and second light sources being associated with separate lighting or signaling functions, and wherein the set of current input data includes a first driving current value of the first light source (103.1) and a second driving current value of the second light source (103.2). A method according to any one of the preceding claims, comprising a prior step of storing (203) the predictive model in a control module (110) of the equipment, the control module being able to access the input data set of the equipment (101). A method according to any one of the preceding claims, wherein the predictive model is obtained by a prior supervised learning step (201) on the basis of training data, the training data comprising a training set of equipment input data corresponding respectively to temperature measurements of the equipment point of interest. A method according to any one of the preceding claims, wherein the predictive model is further applied to a current ambient temperature measurement from an ambient temperature sensor (105). Computer program comprising instructions for implementing the method according to any one of the preceding claims, when these instructions are executed by a processor (301). Control module (110) of a piece of equipment, comprising a memory (302) storing a predictive model from machine learning applied to a training data set; a processor (301) configured to obtain a set of current input data including at least one current input data of the equipment, the current input data being representative of a control parameter of the equipment; apply the predictive model to the set of current input data of the equipment and to at least one current temperature of the equipment; following the application of the predictive model, prediction of a future temperature of the first point of interest of the equipment.

Images