Method, device and system for monitoring a rotating electrical machine

Abstract

A method for monitoring a rotating electrical machine, in particular an electric generator such as a wind turbine generator, device and system. The method comprises the following steps: a step of receiving representative temperature data from a part of the machine; a step of determining a theoretical temperature from a thermal model parameterized by parameters representative of the machine's operation and / or parameters representative of the environment; a step of comparing the received data with the theoretical temperature; and a step of comparing the difference between the received data and the theoretical temperature with a predetermined threshold. Figure for the abstract: Fig. 1

Description

Description Title of the invention: Method, device and monitoring system of a rotating electrical machine General technical field and prior art

[0001] — The present invention relates to the general field of machine monitoring rotating electrical components, for example electric generators which can be present in wind turbines.

[0002] — More particularly, the invention relates to the early detection of defects in rotating electrical machines in operation using measurements, for example measurements stored in a SCADA system.

[0003] — The SCADA system for Data Acquisition and Control System in English "Supervisory Control And Data Acquisition" is a remote management system large scale enabling real-time processing of a large number of telemetry data and to remotely control technical installations.

[0004] — The production of renewable energy from wind power is developing rapidly. While offshore wind farms are in the development and Construction, land-based parks are more in the asset management and Maintenance. Wind turbine generators consist of a set A moving part driven in rotation by the airflow. This moving assembly comprises an electric generator, which is an important component ensuring the conversion of mechanical energy is converted into electrical energy. These generators can exhibit Breakdowns that cause production losses and increased maintenance costs. The most common faults encountered in this type of machine are short-circuit faults. between turns of the stator and rotor windings and power supply faults rotor. To prevent the deterioration of defects and increase the availability of resources By scheduling repair stops in advance, the operator hopes to have a reliable diagnostic tool for operating generators that allows detection Early detection of defects allows for preventive maintenance.

[0005] — Current practices for detecting defects in generators are the Generator standby diagnosis, vibration diagnosis and electrical diagnosis

[0006] — Standby diagnosis is effective for detecting faults but it requires the generator to be shut down the machine and does not allow monitoring of the generator's condition. Thus, it intervenes ge- generally after the defect has worsened and the machine has deteriorated. This type of diagnosis is carried out mainly after the manufacturing and installation phases of the Equipment. The different types of tests for diagnosing stopped machines are: - Voltage withstand tests of stator and rotor windings. - Measurements of capacity, dissipation factor and power factor. - Measurements of partial discharges. - Measurements of insulation resistance, polarization index and dielectric absorption ratio. - Measurement of winding resistance. - Analysis of the frequency sweep response. Vibration diagnostics is the most widely used diagnostic method in wind turbine generators for detecting faults in operating generators and throughout the entire electromotive force (EMF) section of wind turbine generators. Vibration diagnostics involves performing a harmonic analysis of vibration signals measured by sensors installed on the equipment being monitored. The drawback is that it does not allow for the efficient detection of faults. Electrical diagnostics is the most effective online diagnostic method for wind turbine generators, as it provides a precise picture of the machine's condition. It is based on electrical measurements (currents, voltages, power, etc.) taken at a high sampling frequency (tens of kHz) and the harmonic analysis of this data (Fourier transform, wavelet transformation, etc.). The drawback of this diagnostic method is the need to equip the generator with electrical sensors and high-frequency data acquisition tools. This requires a significant investment (equipment and the time of personnel responsible for installing this equipment) to apply the diagnostic method to all wind turbines. Furthermore, this diagnostic process only begins at the time of sensor installation and does not allow for the study of historical data. General presentation of the invention A general aim of the invention is to overcome the limitations of the prior art, in particular by proposing a method of monitoring an electrical machine enabling the planning of maintenance operations and the reduction of maintenance costs. The method and device of this invention can be applied to any rotating electrical machine (motor or alternator). To this end, the invention proposes a method for monitoring a rotating electrical machine, in particular an electric generator, such as a wind turbine generator. The method comprises the following steps: a step of receiving representative temperature data from a part of the machine; a step of determining a theoretical temperature from a thermal model parameterized by parameters representative of the machine's operation and / or parameters representative of the environment; a step of comparing the received data with the theoretical temperature; and a step of comparing the difference between the received data and the theoretical temperature with a pre-defined threshold. determined. The monitoring process can be implemented in the following way. In one embodiment of the process, the machine includes a stator, the stator having at least one phase and a coil associated with the phase. Furthermore, the part of the machine from which the temperature data representative of the phase are extracted is the coil. In one embodiment of the process, the machine includes a stator, the stator having at least three phases and three coils associated respectively with the three phases, and the machine has at least three thermocouples mounted on three heads associated respectively with the three coils. Furthermore, the receiving step is configured to receive data from the three thermocouples, the theoretical temperature determination step is configured to determine three theoretical temperatures from three parameterized thermal models, and the received data comparison step is configured to compare the received data with the three theoretical temperatures. In one embodiment, the process further includes a step of determining the predetermined threshold, this predetermined threshold being a function of temperatures of the part of the machine, taken on a healthy machine and received during a time interval. In one embodiment, the process further includes, before the comparison step, a first step of deleting data that does not represent any value and / or a second step of deleting data received in at least two copies and / or a third step of deleting data with a value less than a minimum and / or data with a value greater than a maximum. In one embodiment, the process further includes a step of determining an average and / or a maximum value and / or a minimum value of the data received during a time interval; moreover, the step of comparing the received data is configured to compare the average and / or the maximum value and / or the minimum value with the theoretical temperature. In one embodiment, the process includes a learning phase during which measurements are taken on a healthy machine to extract and store indicators to calibrate the parameterized thermal model, for example over a period of about one year, and a threshold adjustment phase, and a monitoring phase. In one embodiment of the process, the parameters representative of the operation of the machine include power measurements, in particular active power as a function of torque and / or current measurements and / or voltage measurements and / or rotational speed measurements and / or the parameters representative of the environment include wind speed and / or ambient temperature. The invention also proposes a device for monitoring a rotating electrical machine, in particular an electric generator, such as a wind turbine generator. The device includes an input for receiving representative temperature data from a part of the machine and a data processing unit configured to perform a step of receiving representative temperature data from a part of the machine and a step of determining a theoretical temperature from a thermal model parameterized by parameters representative of the machine's operation and / or parameters representative of the environment and a step of comparing the received data with the theoretical temperature and a step of comparing the difference between the received data and the theoretical temperature with a predetermined threshold. In one embodiment of the monitoring device, the machine includes a stator, this stator includes at least one phase and a coil associated with the phase, and the part of the machine from which representative temperature data are extracted is the coil. In one embodiment of the monitoring device, the machine includes a stator. The stator has at least three phases and three coils associated with each phase. Furthermore, the machine includes at least three thermocouples mounted on three heads, each associated with one of the three coils. The input of the monitoring device is configured to receive data from the three thermocouples. Additionally, the theoretical temperature determination step is configured to determine three theoretical temperatures from three parameterized thermal models, and the received data comparison step is configured to compare the received data with the three theoretical temperatures. The invention also proposes a system comprising a rotating electrical machine, in particular an electric generator, such as a wind turbine generator, and a monitoring device connected to the generator. The monitoring device includes an input for receiving representative temperature data from a part of the machine and a data processing unit configured to perform a step of receiving representative temperature data from a part of the machine, a step of determining a theoretical temperature from a thermal model parameterized by parameters representative of the machine's operation and / or parameters representative of the environment, a step of comparing the received data with the theoretical temperature, and a step of comparing the difference between the received data and the theoretical temperature with a predetermined threshold. In one embodiment of the system, the machine includes a stator; this stator has at least three phases and three coils associated respectively with the three In phases, the machine includes at least three thermocouples mounted on three heads, each associated with one of the three coils. The monitoring device input is configured to receive data from all three thermocouples. Furthermore, the theoretical temperature determination step is configured to determine three theoretical temperatures from three parameterized thermal models, and the received data comparison step is configured to compare the received data with the three theoretical temperatures. In one embodiment the system further includes a database configured to store data. In one embodiment of the system, the database is configured to determine an average and / or a maximum value and / or a minimum value of the data received during a time interval. Furthermore, the data comparison step is configured to compare the average and / or the maximum value and / or the minimum value with the theoretical temperature. In one embodiment of the system, the average and / or maximum value and / or minimum value of the received data is determined with a recurrence period of approximately 10 minutes. Brief description of the drawings Other features and advantages of the invention will become apparent from the following description, which is purely illustrative and not limiting, and should be read in conjunction with the accompanying figures in which: [fig.1] Figure | illustrates a system according to the present invention. [fig.2] Figure 2 illustrates a monitoring method according to the present invention. [fig.3] Figure 3 illustrates learning the thermal behavior of a healthy generator within the framework of the present invention. [fig.4] Figure 4 illustrates an adjustment of the anomaly detection threshold within the framework of the present invention. [fig.5] Figure 5 illustrates online monitoring of the thermal behavior of the generator within the framework of the present invention. [fig.6] Figure 6 illustrates duplicated data. [fig.7] Figure 7 illustrates active power filtering as a function of torque within the framework of the present invention. [fig.8] Figure 8 illustrates a distribution of active power for each wind interval of 0.5m / s. [fig.9] Figure 9 illustrates different pretreatments according to the present invention. [fig.10] Figure 10 illustrates a thermal model of the generator. [fig.11] Figure 11 illustrates a thermal model of the generator and the averaged residual temperature. [fig.12] Figure 12 illustrates the detection of faulty generators on the Salles-Curan data. Description of implementation and execution methods Figure | represents a system 101 according to the present invention. This system 101 comprises: - a rotating electrical machine 102, in particular an electric generator, such as a wind turbine electric generator, and - a monitoring device 103, connected to the generator. The 103 monitoring device includes: - an input 103-a to receive representative temperature data from a part of the machine and - a 103-b data processing unit. The 103-b processing unit is configured to perform a monitoring process as shown in Figure 2. This monitoring process includes: - a step 201 of receiving representative temperature data from a part of the machine and - a step 202 of determining a theoretical temperature from a thermal model parameterized by parameters representative of the machine's operation and / or parameters representative of the environment and - a step 203 comparing the received data with the theoretical temperature and - a step 204 of comparing the difference between the received data and the theoretical temperature with a predetermined threshold. Thus, the monitoring process is based on comparing the measured temperature of a portion of the generator with the temperature calculated from a thermal model. This diagnostic method allows for early (several months before the problem worsens) and effective detection of faults in the operating generator across all operating ranges, using average measurements and minimizing the number of sensors required on the wind turbine. In one embodiment, the monitoring of the electrical machine, for example, by way of illustration, a wind turbine generator, is based on comparing the expected and measured winding temperature of a stator phase of the generator. The expected behavior is derived from a statistical thermal model trained on a healthy wind turbine. It takes into account the operating and environmental parameters of the wind turbine in order to estimate the temperature of a stator winding phase of the generator. The difference between the measured temperature and the prediction of the thermal model, also called the residual temperature, allows for quantify the behavioral deviation. It is then possible to set a threshold on this deviation beyond which the generator will be considered to be faulty. The temperature of the generator's stator windings can be measured, for example, using three thermocouples. Each thermocouple is placed, for instance, on the coil heads of one of the three stator phases and measures the winding temperature of that phase. Other measurements of the wind turbine's operation (power, current, voltage, rotational speed, etc.) and the environment (wind speed, ambient temperature, etc.) are also taken within the turbine. All measurements are transferred and stored in a remotely accessible database. For example, a remote management system called SCADA (Supervisory Control and Data Acquisition) can be used. Advantageously, for each 10-minute period, the average, maximum, or minimum of the observed values ​​are calculated. The monitoring process is, by way of illustration and not limitation, carried out in three phases An illustrated learning phase is shown in figure 3: The thermal model of the generator is trained during the learning phase on a known healthy electrical generator. This statistical model provides a temperature estimate for each phase of the generator, taking into account the various operating parameters. A new model must be generated for each wind turbine and generator technology. As illustrated by element 301, the training phase advantageously requires one year of measurement history on a generator deemed healthy. Once step 302, which involves data cleaning, preprocessing, and filtering, step 303, which extracts indicators, is performed. These indicators are used in step 304 as input to the generator thermal model to calibrate the parameters of the healthy generator's thermal model. The model parameters are then retained to subsequently estimate the expected temperature of each phase of the generator. Threshold adjustment phase illustrated in figure 4: The fault threshold, set on the residual temperature averaged over several time points (for example, as a non-limiting example, 144 measurement points, which is equivalent to 24 hours for 10-minute points), will allow us to distinguish healthy generators from those with faults. Historical data on healthy and faulty generators are necessary to best adjust the threshold on the averaged residual temperature. This historical data is stored in a SCADA-type database. The residual temperature is obtained by the operation 402 of subtracting the calculated temperature from the measured temperature. Then the operation 403 corresponds to a moving average of this difference. Finally, there is a 404 operation for comparison with a threshold. The objective is to detect all thermal faults (few true negatives) while limiting the number of false positives. The value of this threshold will depend on the model and therefore on the wind turbine and generator technologies. Once adjusted, it will be used during the monitoring phase. Monitoring phase illustrated in figure 5: The monitoring phase consists of continuously detecting average residual generator temperatures that exceed the threshold. It will utilize the model parameters and the detection threshold to distinguish between healthy and faulty thermal behavior of the generator. To achieve this, the SCADA-type 501 database will contain the data received continuously from the generator. Applying a filter based on the deviation from the model, rather than implementing an absolute threshold on the measured parameters, allows for more accurate monitoring of the generator's behavior in relation to the wind turbine's operation, thus providing a more effective anomaly detection tool. The following paragraphs illustrate, in a non-exhaustive manner, the processing of data. The data used to perform this diagnostic is SCADA-type data that measures various physical parameters of the wind turbine (temperatures, active and reactive power, currents, voltages, rotational speeds, wind speed, etc.). This data is stored in a database, for example, the PI OSIsoft database. It is also possible to replicate this data, for example, in the REDSHIFT database from Amazon Web Services. The data used is averaged over a 10-minute time step. The quantities retrieved are the powers, voltages of each phase, currents of each phase, temperatures of each phase, the rotation speed of the generator, the ambient temperature, and the wind speed. The first preprocessing step consists of cleaning the measured data, without prior knowledge of the material to be studied. First, duplicate data is processed. For the same electrical generator, the same date, and the same measured quantity, it is possible to have several different values ​​in the database (as illustrated in Figure 6). These represent approximately 2% of the retrieved data volume. For each triplet consisting of electrical generator, date, and measured quantity, we retained the average value of the available non-zero quantities (in the computer science sense, here NaN stands for Not a Number). Next, the null data is processed. The null data consists of the measurements These values, which are not associated with any particular value, represent up to 30% of the data for certain measured quantities. When one of these quantities is zero at a given date, all measurements at that date are eliminated from the study. The second preprocessing step consists of cleaning the data by adding knowledge related to the wind turbine and in particular to its electrical generator. Next, a range of acceptable values ​​is considered. Each measured quantity has ranges of acceptable values, as well as ranges of non-physical values. Some of these acceptable ranges will depend on the wind turbine technology and the various materials that compose it. Thus, filters are applied to measurements of frequency, wind speed, and turbine-dependent measurements such as active power, temperatures, and winding voltage. Consideration of acceptable multi-measurement ranges. Illustration of active power as a function of torque. The active power produced by the wind turbine varies linearly with torque. To eliminate values ​​that deviate from this law, we used a linear regression of the active power data against torque. Measurements that deviate by more than 150 kW from this regression are discarded. Figure 7, which represents the active power as a function of the torque, illustrates this pre-processing, with part 701 showing the data that are not retained after pre-processing, line 702 representing the linear regression, and part 703 showing the measurements retained after pre-processing. The behavior of a wind turbine depends on numerous factors, such as operating conditions, wear and tear on its various components, and so on. To focus solely on the behavior of the electrical generator without being influenced by other phenomena, the data retained are the most accurate representations of the power curve under ideal wind conditions. This selection allows, among other things, the elimination of periods of turbine curtailment, as well as outliers corresponding to periods of unstable operation (due to the use of 10-minute averaged values). For each wind speed range of 0.5 m / s, we retrieve the median value of the active power. This data allows us to reconstruct the theoretical power curve as a function of wind. When the data deviate from this median value by more than 300 kW, the measurements are excluded from the study. Figure 8 illustrates the distribution of active power for each wind range of 0.5m / s with the distribution histogram, the median represented in 801 and different quantiles (0.05, 0.25, 0.75 and 0.95) represented by 801-a to 801-d. The bands represent the ranges of accepted values. Figure 9 illustrates the evolution of the data following the two preprocessing steps based on multi-measurement acceptable ranges. Measurements during the first preprocessing steps are shown in 901, and measurements after the final preprocessing steps are shown in 902. The thermal model used to diagnose the generator takes as input calculated indicators based on SCADA-type measurements. Certain representative parameters allow for the representation of the generator's physical characteristics and the heating at the phase terminals. Other representative parameters characterize the wind turbine's environment, the heating generated during its operation, and the inertia of temperature evolution. Representative environmental parameters include generator phase temperatures which depend on external environmental conditions, particularly outside temperature. The parameters representative of the machine's operation include heating, which is related to the power produced by the wind turbine. At low power, the generator will have a lower temperature than at high power. Furthermore, a higher reactive power will tend to generate more heating within the generator. The generator losses due to the resistance of the windings are proportional to the square of the current. These losses correspond to the heating of the coils as the current passes through them. Furthermore, temperature is a quantity that exhibits inertia. Its value at a given instant depends on the temperature at the previous instant. The model uses the aforementioned powers to account for this inertia. Thus, at each instant, the quantities presented in the table below are calculated or retrieved: [Table 1] Characteristics Calculation active_power | reactive_power_abs | abs(reactive_power_avg) | apparent_power | (active power_ang)* + (reactive power avg)” [loss _generatrice ((turbine_current_phase1_avg)? + (turbine_current_phase2_avg)* + (turbine_current_phase3_avg)?) | *0.001 | average_current (turbine_current_phasel_avg + turbine_current_phase2_avg + turbine_current_phase3_avg) / 3 | | medium_voltage (turbine_voltage_phase1_avg + turbine_voltage phase2_avg + turbine_voltage_phase3_avg) / 3 | generator_speed | ambient_temperature wind speed |previous_active_power active_power_avg [t — 1] |average_active_power | mean(active power averagedt - 3:t]) The statistical model used is a multivariate linear regression taking into account all these indicators in order to estimate a theoretical phase temperature. The entire section concerning the generator's behavior is based on physical knowledge and the potentially nonlinear laws integrated into the calculated indicators; see in particular Figure 10. In this figure, the input parameters that allow the use of the generator's thermal model are active power, reactive power, apparent power, generator losses, average currents, average of the three voltages, generator speed, outside temperature, wind speed, previous active power, and the average of the active powers. The model learns the behavior of a healthy generator over a period that, advantageously, exceeds one year. This allows for a sufficient variety of input data to the model, reducing the discrepancy between the values ​​calculated by the model and the measured values, and thus making the model as close to reality as possible. By applying the learning curve from a healthy wind turbine over one year and evaluating it on other healthy wind turbines, the average error in the estimation is less than 3%. This estimation performance will allow for setting a sufficiently precise threshold to detect abnormal heating of the generator. The comparison between the measured temperature and the expected temperature calculated by the thermal model yields the residual temperature. This residual temperature is then averaged over a sliding window of 144 points, corresponding to one day, to smooth the residual signal before comparison with the fault threshold. The average residual temperature helps avoid false positives (false alarms) and provides greater certainty in the diagnostic result. In Figure 11, this comparison and filtering are performed by elements 1101 and 1102. The approach presented is applied simultaneously to the three stator phases of the generator. Three averaged residual temperatures are then compared to the threshold. This provides a better understanding of the generator's condition and makes it possible to locate the source of the fault within the generator (stator, rotor, or cooling system). Once the three models are adjusted, they can be used to calculate the predicted temperature of each of the three phases at any given time. This calculated temperature is compared to the measured temperature phase by phase to obtain the averaged residual signal. By setting a threshold on this averaged residual temperature, it is then possible to detect atypical generator heating behaviors. To validate the diagnostic and detection method, historical data from healthy and faulty generators were retrieved from the Salles-Curan site. The thermal model was generated on one of the healthy generators. By replaying this method on the historical data, faulty generators are clearly differentiated from healthy ones, and with this model, by setting a threshold between 12 and 14, it is possible to detect the occurrence of the fault early in order to anticipate maintenance operations, as shown in Figure 12. Following the detection of a malfunction, it is possible to identify its cause. If only one or two of the three phases are out of phase, the fault likely originates in the stator. If all three phases are affected, the fault may concern either the rotor or the cooling system. To delve deeper into the diagnosis of the If the problem persists, the cooling system temperature should be checked to determine whether or not the cooling system is the issue. If the cooling temperature is significantly higher than the manufacturer's specification, the problem likely lies within the cooling system itself.

Claims

Demands

1. Method for monitoring a rotating electrical machine (102), including an electric generator, such as a generator Wind turbine electricity, comprising the following steps: - a step (201) of receiving representative data of temp- temperatures of a part of the machine and - a step of (202) determining a theoretical temperature from of a thermal model parameterized by parameters representative of the machine operation and / or representative parameters the environment and - a step (203) of comparing the received data with the tem- theoretical and - a step (204) of comparing the difference between the data received and the theoretical temperature with a predetermined threshold.

2. A monitoring method according to claim 1, wherein: - the machine includes a stator, -the stator has at least one phase and an associated coil phase ; - the part of the machine from which the representative data is extracted The coil is the temperature sensor.

3. A monitoring method according to claim 1 or 2, wherein: - the machine includes a stator, - the stator comprises at least three phases and three associated coils res- with respect to the three phases, - the machine includes at least three thermocouples placed on three heads associated respectively with the three reels; - the reception step (201) is configured to receive data from three thermocouples; - the step of (202) determining a theoretical temperature is configured for determining three theoretical temperatures at starting from three parameterized thermal models and - the step (203) comparison of received data is configured for the comparison of the data received with the three temperatures theoretical.

4. A monitoring method according to any one of claims 1 to 3 comprising furthermore : - a step of determining the predetermined threshold, this pre- determined as a function of the temperatures of the machine part, taken from a healthy machine and received during a time interval.

5. A monitoring method according to any one of claims 1 to 4 comprising Furthermore, before the comparison stage: - a first step of deleting data that does not represent any value and / or - a second step of deleting the data received in at least two copies and / or - a third step of deleting data of a lower value to a minimum and / or data with a value greater than one maximum.

6. A monitoring method according to any one of claims 1 to 5 comprising furthermore, - a step of determining an average and / or a value maximum and / or minimum value of the data received during an in- temporal interval; moreover ; - step (203) of comparing received data is configured for compare the average and / or the maximum value and / or the minimum value with the theoretical temperature.

7. A monitoring method according to any one of claims 1 to 6 including: - a learning phase during which measurements are taken on a healthy machine to extract and store indicators allowing calibration of the parameterized thermal model, for example on a period of approximately one year, and - a threshold adjustment phase, and - a monitoring phase.

8. A monitoring method according to any one of claims 1 to 7 wherein the parameters representative of the machine's operation com- take measures, particularly active power in depending on the torque and / or current measurements and / or measurements of voltages and / or rotational speed measurements and / or The representative environmental parameters include a wind speed and / or ambient temperature.

9. Device (103) for monitoring a rotating electrical machine (102), in particular an electric generator, such as a generator wind turbine electric, comprising: - an input (103-a) for receiving representative data temperatures of a part of the machine and - a data processing unit (103-b) configured to perform: - a step (201) of receiving representative data of temp- temperatures of a part of the machine and - a step (202) of determining a theoretical temperature from of a thermal model parameterized by parameters representative of the machine operation and / or representative parameters the environment and - a step (203) of comparing the received data with the tem- theoretical and - a step (204) of comparing the difference between the data received and the theoretical temperature with a predetermined threshold.

10. A monitoring device according to claim 9, wherein: - the machine includes a stator, - the stator comprises at least one phase and an associated coil phase ; - the part of the machine from which the representative data is extracted The coil is the temperature sensor.

11. Monitoring device (103) according to claim 9 or 10 in which : - the machine (102) includes a stator, - the stator comprises at least three phases and three associated coils res- with respect to the three phases, - the machine includes at least three thermocouples placed on three heads associated respectively with the three reels; - The input of the monitoring device is configured to receive data from the three thermocouples; - step (202) of determining a theoretical temperature is configured for determining three theoretical temperatures at starting from three parameterized thermal models and - step (203) of comparing received data is configured for the comparison of the data received with the three temperatures theoretical.

12. System comprising: - a rotating electrical machine (102), in particular a generator electrical, such as a wind turbine generator, and - a monitoring device (103), according to any one of claims 9 to 11, connected to the generator; the monitoring system comprising: - an input (103-a) to receive representative data of tem- temperatures of a part of the machine and - a data processing unit (103-b) configured to perform: - a step (201) of receiving representative data of temp- temperatures of a part of the machine and - a step (202) of determining a theoretical temperature from of a thermal model parameterized by parameters representative of the machine operation and / or representative parameters the environment and - a step (203) of comparing the received data with the tem- theoretical and - a step (204) of comparing the difference between the data received and the theoretical temperature with a predetermined threshold.

13. System according to claim 12 in which: - the machine includes a stator, - the stator comprises at least three phases and three associated coils res- with respect to the three phases, - the machine includes at least three thermocouples placed on three heads associated respectively with the three reels; - the input (103-a) of the monitoring device is configured for receive data from the three thermocouples; - step (202) of determining a theoretical temperature is configured for determining three theoretical temperatures at starting from three parameterized thermal models and - step (203) of comparing received data is configured for the comparison of the data received with the three temperatures theoretical.

14. A system according to claim 12 or 13 further comprising a base data configured to store data.

15. System according to claim 14 in which: - the database is configured to determine an average and / or a maximum and / or a minimum value of the data received during a time interval; - step (203) of comparing received data is configured for compare the average and / or the maximum value and / or the minimum value with the theoretical temperature.

16. System according to claim 15 wherein the average and / or the The maximum value and / or the minimum value of the received data is determined. completed with a recurrence period of approximately 10 minutes.