Motor controllers and methods for estimating motor shaft angle and / or motor shaft rotation speed.

By using a redundantly configured motor controller and automatic encoder network, the sensor failure problem in motor position and speed estimation is solved, enabling accurate and reliable motor shaft angle and speed estimation in high-safety applications, ensuring system stability and safety.

CN123074346APending Publication Date: 2026-11-10INFINEON TECHNOLOGIES AG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610614722.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2025-05-08
Filing Date
2026-05-07
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In the prior art, the estimation of motor position and speed is easily affected by sensor failure, leading to unacceptable errors, especially in applications with high safety requirements, making it difficult to achieve accurate and safe motor position and speed estimation.

Method used

The redundant motor controller uses a feedback loop and an automatic encoder network to feed back the angle estimation error signal. It utilizes automatic encoder training and anomaly detectors to detect faults, thereby achieving accurate estimation of the motor shaft angle and speed, and performing safety actions in case of faults.

Benefits of technology

It improves the accuracy and safety of motor position and speed estimation, can detect and respond to sensor failures, and ensures stable operation of the system in high-reliability and safety applications.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123074346A_ABST
    Figure CN123074346A_ABST
Patent Text Reader

Abstract

Motor controllers and methods for estimating motor shaft angles and / or motor shaft rotational speeds of a motor. According to various embodiments, a motor controller is described comprising an estimation circuit and an anomaly detector. The estimation circuit is configured to perform motor shaft angle estimation using a feedback loop, the feedback loop including feedback of an angle estimation error signal generated by the estimation circuit. The anomaly detector includes an autoencoder trained on segments of the angle estimation error signal, wherein the anomaly detector is configured to supply the autoencoder with segments of the angle estimation error signal generated by the estimation circuit, determine reconstruction errors of the autoencoder for those segments, and perform safety actions based on values ​​derived from the reconstruction errors.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to a motor controller and method for estimating the motor shaft angle and / or the motor shaft rotation speed of an electric motor. Background Technology

[0002] In electric motor control, accurate estimation of motor position and speed is crucial for ensuring optimal performance and safety. Motor position and speed estimation is typically accomplished based on signals provided by motor sensors, such as resolvers. These sensors are robust electromechanical devices that provide precise angular position feedback, making them ideal for harsh environments and high-reliability applications. However, errors such as short circuits or open circuits can still occur, leading to position or speed failures, which may be unacceptable depending on safety requirements. Therefore, methods for accurate and safe motor position and / or speed estimation are desirable. Summary of the Invention

[0003] According to various embodiments, a motor controller is provided, including an estimation circuit and an anomaly detector. The estimation circuit is configured to perform motor shaft angle estimation using a feedback loop, the feedback loop including feedback of an angle estimation error signal generated by the estimation circuit. The anomaly detector includes an autoencoder trained on a segment of the angle estimation error signal, wherein the anomaly detector is configured to supply the autoencoder with the segment of the angle estimation error signal generated by the estimation circuit, determine the reconstruction error of the autoencoder for the segment of the angle estimation error signal, and perform a safety action based on a value derived from the reconstruction error.

[0004] According to another embodiment, a method is provided for estimating the motor shaft angle and / or motor shaft rotation speed of a motor based on the above motor controller. Attached Figure Description

[0005] In the accompanying drawings, similar reference characters generally refer to the same parts throughout the different views. The drawings are not necessarily drawn to scale; instead, the focus is generally on illustrating the principles of the invention. In the following description, various aspects are described with reference to the following drawings, wherein: Figure 1 A motor control arrangement according to an embodiment is shown.

[0006] Figure 2 A rotary transformer-to-digital converter according to an embodiment is shown.

[0007] Figure 3 A motor control arrangement according to an embodiment is shown.

[0008] Figure 4 An autoencoder network according to an embodiment is shown.

[0009] Figure 5 The diagram illustrates the expected signal behavior.

[0010] Figure 6 The diagram illustrates the signal under nominal conditions.

[0011] Figure 7 The diagram illustrates the signal under conditions of a short circuit at the output of the rotary transformer.

[0012] Figure 8 The diagram illustrates the signal when the input is disconnected.

[0013] Figure 9 The diagram illustrates the signal when the output signal of the rotary transformer is rectangular.

[0014] Figure 10 The diagram illustrates the signal when the output signal of the rotary transformer has been reduced in amplitude.

[0015] Figure 11 The diagram illustrates the output signal of the resolver after the amplitude has been reduced using the sliding window method.

[0016] Figure 12 A motor controller according to an embodiment is shown.

[0017] Figure 13 A flowchart illustrating a method for estimating the motor shaft angle and / or the motor shaft rotation speed is shown. Detailed Implementation

[0018] The following detailed description refers to the accompanying drawings, which illustrate by way of example specific details and aspects in which the present disclosure may be practiced. Other aspects may be utilized, and structural, logical, and electrical changes may be made, without departing from the scope of the invention. The various aspects of this disclosure are not necessarily mutually exclusive, as some aspects of this disclosure may be combined with one or more other aspects of this disclosure to form new aspects.

[0019] Figure 1 A motor control arrangement 100 according to an embodiment is shown.

[0020] The rotor 101 of the resolver 102 is attached to the motor shaft driven by the motor 103 (e.g., a permanent magnet synchronous motor (PMSM)) such that the angular position (i.e., angle) and rotational speed of the resolver rotor 101 correspond to the motor rotational speed and the angular position of the motor rotor 104.

[0021] The motor 103 is controlled by a microcontroller 105 (operating as a motor controller), which includes a resolver-to-digital converter 106 (receiving the output signal from the resolver 102), a carrier signal generator 107 (supplying a carrier signal to the resolver rotor 101), and a PWM (pulse width modulation) generator 108.

[0022] During operation, carrier signal generator 107 generates a carrier signal for output to resolver 102. Resolver 102 is an electrical transformer that measures the rotation angle of the motor shaft driven by motor 103. Resolver 102 has three coil windings: a primary winding supplied with a carrier signal and two secondary windings that output the measurement signal, referred to as a sine and cosine signal (although strictly speaking, they are carriers with sine and cosine envelopes, respectively). The secondary windings are mounted 90° apart from each other relative to the motor shaft. The primary winding is mounted on resolver rotor 101 (and thus on the motor shaft) and is excited by the carrier signal applied to it during operation. As the motor shaft rotates, the excited primary winding induces a sinusoidally changing voltage in the secondary windings, thereby generating the sine and cosine signals. The sine and cosine signals are amplitude-modulated versions of the carrier signal, each having (ideally) a sine and cosine waveform as its envelope, respectively. The rotary transformer-to-digital converter 106 uses sine and cosine signals to determine the angular position θ of the motor shaft (according to sine θ / cosine θ = tangent θ), and determines the rotational speed of the motor shaft by taking into account the change of angular position over time.

[0023] The microcontroller 105 controls the motor via PWM, that is, by controlling the PWM generator 108 to output a PWM signal according to the desired acceleration, and supplying the PWM signal to the motor (e.g., via the eGTM (enhanced gate driver module) to supply the IGBT (insulated gate bipolar transistor) hybrid pack of the motor) to control the motor accordingly.

[0024] The microcontroller 105 can, for example, execute space vector PWM. This can include field-oriented control (FOC) using measurements of the average current through the motor windings as input (in addition to angle and speed information provided by the resolver-to-digital converter 106).

[0025] Therefore, the motor control arrangement 100 forms a control loop for controlling the motor according to closed-loop control, wherein the control signals (PWM signals for the three phases of the motor) are determined by a microcontroller (i.e., the motor controller) based on measurement results (sine and cosine signals and possibly average current measurements), and are thus adapted to the feedback path (providing the measurement results). An example of the structure and functionality of the resolver-to-digital converter 106 is described below.

[0026] Figure 2 A rotary transformer-to-digital converter 201 according to an embodiment is shown.

[0027] The resolver-to-digital converter 201 can be implemented in software, for example, on the CPU of the microcontroller 105 or on the digital signal processor (e.g., CDSP (converter digital signal processor)) of the microcontroller 105.

[0028] The resolver-to-digital converter 201 includes a feedback (FB) filter chain loop 202 that provides speed and angle estimates, and optionally includes (e.g., to a CPU performing PWM control, as referenced above) Figure 1 As described, a post-processor performs post-processing before outputting angle and velocity estimates.

[0029] The FB filter chain loop 202 includes a pair of analog-to-digital converters 203, 204 (e.g., delta-sigma analog-to-digital converters (DS-ADCs)) that provide digital versions of the sine and cosine signals provided by the resolver 205 (corresponding to resolver 102) (i.e., time series of 16-bit samples of the sine and cosine signals, respectively).

[0030] Optionally, the digitized cosine signal can be processed by the first sinc filter 208 before being supplied to the first mixer 206, and the digitized sine signal can be processed by the second sinc filter 209 and the gain adjuster (multiplier) 210 at a certain gain before being supplied to the second mixer 207.

[0031] Therefore, this pair of DS-ADCs is configured to draw from the reference above. Figure 1 The described resolver 205 receives a pair of input sine and cosine signals VSIN and VCOS, which are converted from analog to digital. As mentioned above, these input sine and cosine signals VSIN and VCOS are amplitude-modulated (AM) versions of the carrier signal VCS supplied by the carrier signal generator 211, representing the rotation angle θ of the motor shaft. These input signals may be noisy. In particular, they should be phase-shifted by 90 degrees relative to each other, but this may not be the case due to noise.

[0032] (Optional) sinc filter pairs 208 and 209 are configured to filter digital input sine and cosine signals at VSIN and VCOS, respectively. A sinc filter is a type of low-pass filter that removes high-frequency components above a given cutoff frequency.

[0033] The first mixer 206 and the second mixer 207 are part of the state observer (e.g., Luenberger observer) 212 of the FB filter chain loop 202. The state observer 212 is configured to estimate the rotational speed and angular position (i.e., rotation angle) of the motor shaft based on the digitized (and possibly filtered and gain-adjusted) sine and cosine signals supplied to it.

[0034] The state observer 212 includes a rotational velocity integrator 213 (configured to integrate rotational acceleration to provide a rotational velocity estimate), a first adder 214, a rotational angle integrator 215 (configured to integrate rotational velocity to provide an angle estimate), a sine lookup table (LUT) 216, and a cosine LUT 217, namely a lookup table containing sine values ​​(sine signal value lookup table) and a lookup table containing cosine signal values ​​(cosine signal value lookup table), which can also be combined into a single lookup table.

[0035] The rotational speed integrator 213 is configured to integrate the estimated rotational angle error signal Δθ (indicating the angle estimation error, hereinafter also referred to as the angle error) of the motor shaft and output a rotational speed estimate. The rotational speed integrator 213 integrates its input over the carrier period to generate an envelope signal for subsequent processing.

[0036] The first adder 214 is coupled between the rotational speed integrator 213 and the rotational angle integrator 215. It is configured to add the estimated rotational speed to a damped version of the estimated rotational angle error signal Δθ. .

[0037] The rotation angle integrator 215 is configured to integrate the offset version of the estimated rotational speed provided by the first adder 214 and output the estimated rotation angle of the motor shaft.

[0038] Sine LUT 216 is configured to determine the feedback sine signal VSIN_FB based on the estimated rotation angle, and cosine LUT 217 is configured to determine the feedback cosine signal VCOS_FB based on the estimated rotation angle. First mixer 206 mixes the FB sine signal with the input cosine signal. Second mixer 207 mixes the FB cosine signal with the input sine signal. The outputs of the two mixers 206 and 207 are subtracted from each other by subtractor 218 to generate a (difference) signal representing the estimated rotation angle error signal Δθ, which can be considered as the difference between the state observer's estimate and the actual motor shaft angular position.

[0039] Then, demodulator 219 demodulates the difference signal by multiplying it with the carrier signal VCS to output a demodulated version of the difference signal, namely, an estimated rotation angle error signal Δθ. (It should be noted that in this context, the small angle error Δθ can be approximated as a sine wave (Δθ), which is the component produced by demodulation, and integrators 213 and 215 filter out higher frequency components).

[0040] Then, the estimated rotation angle error signal is gain-adjusted using two constants: the first multiplier 220 multiplies the estimated rotation angle error signal Δθ by ω. o 2 The second adder 221 combines the gain-adjusted version of the rotational speed error signal with the measured acceleration signal vlo to output a gain-adjusted version of the rotational speed error signal to the rotational speed integrator 213, thereby closing the loop. The second multiplier 222 multiplies the estimated rotational angle error signal Δθ by the damping constant. To generate a damped version of the estimated rotation angle error signal Δθ .

[0041] For example, the resolver-to-digital converter elements to the left of sinc filters 208 and 209 are implemented in software, while sinc filters 208 and 209 and DS-ADCs 203 and 204 are implemented in hardware.

[0042] In safety-related applications (such as those requiring high automotive safety integrity levels, such as ASIL-D), it should be possible to avoid or at least detect a failure of the resolver-to-digital converter 106. One approach is to provide its functionality twice in a redundant manner within the microcontroller 105. For example, the microcontroller 105 could implement two state observers 212 (downstream of ADCs 203, 204, sinc filters 208, 209, and gain adjuster 210), and they would operate in parallel, each generating position and velocity estimates. The microcontroller 105 would further implement a comparator that compares the estimates from the two state observers 212 and performs safety actions, such as triggering an alarm in case of a mismatch. However, for example, if one of the ADCs 203, 204 provides a signal that is always zero, then the two state observers 212 receive the same input signal and thus generate the same results, but the position and velocity estimates would still be incorrect.

[0043] To enable the detection of this type of fault, various embodiments describe methods for estimating motor shaft angle and / or motor shaft rotation speed, including a forwarding path for an input signal, which is compared with a signal generated by an estimation circuit, as described in more detail below.

[0044] Figure 3 A motor control arrangement 300 according to an embodiment is shown.

[0045] As in the motor control arrangement 100, the rotor 301 of the rotary transformer 302 is attached to the motor shaft driven by the motor. Figure 3 Not shown in the image, PWM control is also not present. Figure 3 As shown in the figure, the angular position (i.e. angle) and rotational speed of the rotary transformer rotor 301 correspond to the motor rotational speed and the angular position of the motor rotor.

[0046] The microcontroller 305 includes two ADCs 306 and 307 corresponding to analog-to-digital converters (ADCs) 203 and 204, which provide digital versions of the signal output by the resolver 302 (which is generated by the resolver by modulating a carrier signal supplied to the rotor 301 by the carrier generator 304 and the amplifier 303) as described above.

[0047] The microcontroller 305 then includes two estimation paths, implemented in this example by a first estimation circuit 308 and a second estimation circuit 309, respectively. See reference... Figure 2 As described, each estimation circuit 308, 309 implements a state observer 212 (e.g., preceded by sinc filters 208, 209 and gain adjuster 210).

[0048] Therefore, both estimation circuits 308 and 309 generate a velocity estimation signal (i.e., a time series of velocity estimation), an angle estimation (including a time series of angle estimation), and a rotation angle error signal Δθ (including a time series of angle error estimation). Each of these time series includes a corresponding signal value (velocity estimate, angle estimate, and rotation angle error, respectively) for each of the sequences at the estimation time (e.g., the sampling time).

[0049] The microcontroller 305 includes a comparator 310 that compares the speed estimate generated by the first estimation circuit 308 with the speed estimate generated by the second estimation circuit 309, the angle error estimate generated by the first estimation circuit 308 with the angle error estimate generated by the second estimation circuit 309, and / or the angle error generated by the first estimation circuit 308 with the angle error generated by the second estimation circuit 309. If a mismatch is detected, it triggers a safety action.

[0050] Therefore, for safety (e.g., to meet ASIL-D requirements), the microcontroller 305 includes a method in which it provides two estimation functionalities in a redundant manner: estimation circuits 308 and 309 operate in parallel, and comparator 310 compares their results and performs a safety action in case of a mismatch. This increases safety. However, for example, if the input to the estimation is corrupted, such as one of the ADCs 306 and 307 providing a signal that is always zero, then the two estimation circuits 308 and 309 then receive the same input signal and thus produce the same result, but the position and velocity estimation still will not work correctly.

[0051] To cover this type of fault, the microcontroller 305 further includes an angle error anomaly detector 311. The anomaly detector 311, for example, is part of a safety circuit 313 along with a comparator 310. The angle error anomaly detector 311 implements a neural network with an autoencoder architecture (i.e., autoencoder 312) to generate a reconstruction of the angle error input to it (from one of the estimation circuits 308, 309). The angle error anomaly detector 311 then compares the reconstruction with the original angle error (i.e., the already reconstructed one), and treats large reconstruction errors (or multiple large reconstruction errors) as an indication of an anomaly (i.e., a fault), and performs a safety action, such as triggering an alarm.

[0052] Figure 4 An auto encoder network 400, for example used as an auto encoder 312, is shown.

[0053] The auto encoder 400 includes an input layer 401, which consists of nodes (shown as circles) interconnected (e.g., fully connected in this example) with nodes forming a first hidden layer 402.

[0054] Input layer 401 includes nodes (e.g., 32 nodes) that receive input data, where each node receives the value of a segment of the time series (corresponding to the length of the number of nodes in the input layer). For example, the topmost node obtains the current value of the time series, the second node from the top obtains the previous value of the time series, and so on, such that for each estimation time (i.e., each estimation loop performed by estimation circuits 308, 309), the most recent time series value (within, for example, an observation window of length 32) is fed into machine learning model 400. When used as an autoencoder 312, each value fed to the autoencoder is the angle error for the corresponding estimation time. Thus, for example, the topmost node obtains the current angle error, which is the angle error last determined by estimation circuits 308, 309. It should be noted that, in the absence of faults, estimation circuits 308, 309 determine the same angle error, and therefore the angle error can be obtained from either of them.

[0055] The machine learning model 400 has an autoencoder structure (i.e., an autoencoder architecture), where, in this example, the encoder part has only a single hidden layer 402, i.e., the output of the hidden layer 402 is a latent representation 403 of the input data in the latent space. The latent representation 403 is compressed relative to the input data (i.e., it is a vector of fewer values ​​(e.g., 16 values) than the values ​​of the input data (e.g., 32 values) and can be viewed as encoded data (or an encoding of the input).

[0056] It should be noted that 32 input values ​​are just an example, and more input values ​​can be used, such as 201 input values. In such a case, the potential representation 403 could have, for example, 50 or 100 values.

[0057] exist Figure 4 In the example, the decoder also consists of only a single hidden layer (second hidden layer 404) and an output layer 405 (as in the fully connected encoder section). The second hidden layer 404 and the output layer 405 decode the encoded data.

[0058] Automatic encoders may also have more than one hidden layer on the encoder side and / or decoder side.

[0059] The autoencoder is trained to reproduce the input data, meaning the output data should match the input data. Training examples can be generated by obtaining a time series of the angle errors generated by estimation circuits 308 and 309 when no faults occur. The autoencoder can then be trained by batch-feeding segments of the angle error time series (i.e., segments of the angle estimation error signal) to the autoencoder, calculating the loss between that segment and its reconstruction output by the autoencoder for each segment, generating a total loss by summing the losses, and adjusting the autoencoder (i.e., its weights) to reduce the loss (through backpropagation of the total loss).

[0060] In the anomaly detection operation, the value provided by the decoder of the autoencoder 312 for the current input value is used (i.e., if input and output nodes are used in the same order, such as from top to bottom, then the topmost value). Therefore, in Figure 3 In the example, the autoencoder 312 provides a vector of output values ​​(reconstructed angle error), and the vector corresponding to the most recent value in the input vector (e.g., the angle error output by the autoencoder 301 at the topmost output node) is used as the reconstruction for the current anomaly detection time.

[0061] It should be noted that the anomaly detector 311 can perform encoding and reconstruction for each of the anomaly detection time sequences, which can correspond to the estimation loops of the estimation circuits 308 and 309 (i.e., each iteration of the angle estimation), but can also be a subset thereof. That is, the anomaly detector 311 can operate on the angle estimation error generated for the anomaly detection time corresponding to the downsampling of the angle estimation error generated at all estimation times.

[0062] For each anomaly detection time, the anomaly detector 311 compares the angle value for that time with its reconstruction. It can use this individual reconstruction error to decide whether to take a safety action, but it can also aggregate reconstruction errors. For example, the anomaly detector 311 performs comparisons of the input angle error and the reconstructed angle error over multiple time steps (i.e., a window of the anomaly detection time), generating the difference between the input angle error and the reconstructed angle error (reconstruction error), and if the sum of the differences over the window exceeds a threshold, it takes a safety action, such as triggering an alarm. In other words, the anomaly detector can include post-processing of the reconstruction error via a flight integrator (i.e., an integrator accumulating over a sliding window) to detect long-duration anomalies. This helps detect anomalies that include angle errors that are low but persist for relatively long periods.

[0063] Figure 5 The diagram illustrates the expected signal behavior of the angle error.

[0064] Figure 501 illustrates the output of a rotary transformer as a "sine signal" and a "cosine signal" (i.e., a carrier signal with a sine envelope and a cosine envelope).

[0065] Figure 502 illustrates the angle error as determined by estimation circuits 308 and 309.

[0066] Figure 503 shows the speed of the motor as a reference. In this example, there is a speed change 504 at a specific point in time.

[0067] As can be seen, when the speed change 504 occurs, an angle error spike 505 appears. Then, between speed changes 504, these angle errors decrease to near-zero angle error values ​​in steady-state mode (with small fluctuations depending on the operation of the corresponding control loop).

[0068] The autoencoder 312 is trained by estimating the time series of angle errors generated by circuits 308 and 309 when no fault occurs. The autoencoder learns to correct this type of error signal pattern (i.e., reduce the angle error spike 505 to a steady-state mode).

[0069] Figure 6The diagram illustrates the signals under nominal conditions (i.e., the "good" condition with no faults).

[0070] Figure 601 illustrates the output of a rotary transformer as a "sine signal" and a "cosine signal" (i.e., a carrier signal with both sine and cosine envelopes).

[0071] Figure 602 illustrates the angle error and reconstruction (shown in shaded areas) as determined by estimation circuits 308 and 309.

[0072] Figure 603 illustrates the reconstruction error and the value of the flag set by the anomaly detector 311 when an anomaly is detected. In this example, anomaly detection is based on the reconstruction error over a single time step (i.e., comparing the reconstruction error at a single anomaly detection time with a threshold to determine whether to set a flag). Because the signals are as they should be, and the reconstruction error is therefore low, the anomaly flag is always zero (because the threshold is not exceeded).

[0073] Figure 604 shows the speed of the motor as a reference. In this example, there is a speed change 605 at a certain point in time. This figure is omitted in the following figures because it remains constant.

[0074] Figure 7 The diagram illustrates the signal under the condition of a short circuit in the cosine signal output from the rotary transformer.

[0075] Figure 701 illustrates the output sinusoidal signal and cosine signal (i.e., carrier signal with sinusoidal and cosine envelopes) of a resolver. The output cosine signal of the resolver is equal to zero.

[0076] Figure 702 illustrates the angle error and reconstruction (shown in shaded areas) as determined by estimation circuits 308 and 309.

[0077] Figure 703 illustrates the reconstruction error and the value of the flag set by the anomaly detector 311 when an anomaly is detected. In this example, anomaly detection is based on the reconstruction error over a single time step (i.e., the reconstruction error for a single anomaly detection time is compared to a threshold to determine whether to set a flag).

[0078] As can be seen, for multiple anomaly detection times, the reconstruction error exceeded the threshold (it should be noted that the reconstruction error is greater than...). Figure 6 (The scaling is stronger). Therefore, the anomaly detector 311 sets a flag and detects the fault.

[0079] Figure 8 The diagram illustrates the signal received by the microcontroller for the cosine output of the rotary transformer when the input is disconnected and therefore the cosine is held at a fixed voltage.

[0080] Figure 801 illustrates the output sine and cosine signals of a resolver (i.e., carrier signals with sine and cosine envelopes). The output cosine signal of the resolver is equal to 3V.

[0081] Figure 802 illustrates the angle error and reconstruction (shown in shaded areas) as determined by estimation circuits 308 and 309.

[0082] Figure 803 illustrates the reconstruction error and the value of the flag set by the anomaly detector 311 when an anomaly is detected. In this example, anomaly detection is based on the reconstruction error over a single time step (i.e., the reconstruction error for a single anomaly detection time is compared to a threshold to determine whether to set a flag).

[0083] As can be seen, the reconstruction error exceeded the threshold consecutively after a short period of time. Therefore, the anomaly detector 311 set a flag and detected the fault.

[0084] Figure 9 The diagram illustrates the signal when the output sinusoidal signal from the rotary transformer is rectangular.

[0085] Figure 901 illustrates the output sinusoidal signal and cosine signal of a resolver (i.e., carrier signals with sinusoidal and cosine envelopes). The output sinusoidal signal of the resolver is rectangular, meaning the amplitude of the modulated carrier is fixed and does not follow a sinusoidal envelope.

[0086] Figure 902 illustrates the angle error and reconstruction (shown in shaded areas) as determined by estimation circuits 308 and 309.

[0087] Figure 903 illustrates the reconstruction error and the value of the flag set by the anomaly detector 311 when an anomaly is detected. In this example, anomaly detection is based on the reconstruction error over a single time step (i.e., the reconstruction error for a single anomaly detection time is compared to a threshold to determine whether to set a flag).

[0088] As can be seen, the reconstruction error continuously exceeds the threshold after a short period of time, with only minor interruptions. Therefore, the anomaly detector 311 sets a flag and detects the fault.

[0089] Figure 10 The diagram illustrates the signal where the amplitude of the sine and cosine outputs of the rotary transformer has been reduced.

[0090] Figure 1001 illustrates the output sine signal and the output cosine signal of the rotary transformer (i.e., carrier signals with sine and cosine envelopes).

[0091] Figure 1002 illustrates the angle error and reconstruction (shown in shaded areas) as determined by estimation circuits 308 and 309.

[0092] Figure 1003 illustrates the reconstruction error and the value of the flag set by the anomaly detector 311 when an anomaly is detected. In this example, anomaly detection is based on the reconstruction error over a single time step (i.e., the reconstruction error for a single anomaly detection time is compared to a threshold to determine whether to set a flag).

[0093] As can be seen, the reconstruction error does not exceed the threshold because the angle error signal is scaled relative to the nominal condition, and therefore the automatic encoder 312 can reconstruct it with a small reconstruction error. Therefore, the anomaly detector 311 is unable to detect the fault.

[0094] In order to detect this type of fault (reduced magnitude), according to various embodiments, the anomaly detector 311 uses the flying window method as described above, rather than performing anomaly detection based on a comparison of the reconstruction error over a single anomaly detection time with a threshold: the anomaly detector 311 detects anomalies as a reaction to the reconstruction error aggregated over the sliding window (including the current anomaly detection time and a predetermined number of previous anomaly detection times) exceeding a threshold. For example, the sliding window covers hundreds (e.g., 600) detection times.

[0095] This is Figure 11 The diagram in the middle is shown.

[0096] Figure 11 The diagram illustrates the signal where the amplitude of the sine and cosine outputs of the rotary transformer has been reduced.

[0097] like Figure 11 As shown in the figure, the first figure 1101 illustrates the sine signal and the cosine signal output by the rotary transformer, while the second figure 1102 illustrates the angle error and reconstruction (shown in shaded areas) as determined by the estimation circuits 308 and 309.

[0098] Figure 1103 now illustrates the reconstruction error of the aggregation and the value of the flag set by the anomaly detector 311 when an anomaly is detected.

[0099] As can be seen, the reconstruction error of the aggregated data repeatedly exceeds the threshold. Therefore, the anomaly detector 311 sets a flag and detects the fault. Using an anomaly detection method based on the aggregated reconstruction error, the anomaly detector 311 is thus able to detect the fault.

[0100] Anomalies can be detected because, for smaller input signals, the settling time (from the peak of the reconstruction error) is prolonged, and therefore the reconstruction error accumulates more strongly during that time than it would under nominal conditions.

[0101] It should be noted that the two detections can be combined: the anomaly detector 311 can compare individual reconstruction errors with a reconstruction error threshold, and compare the aggregation of reconstruction errors with an aggregation threshold.

[0102] The aggregated reconstruction error (i.e. the sum of reconstruction errors) can be viewed as a value derived from all the individual reconstruction errors (which fall into a sliding window based on their position at the anomaly detection time).

[0103] In the example described above, as described above, the resolver 302 provides a carrier signal modulated by (ideally, especially in the absence of faults) sine and cosine envelopes as a signal representing the position of the motor over time. However, in other embodiments, a motor sensor that directly provides sine and cosine signals can also be used. In that case, the demodulator 219 is not necessary (or can be operated with a constant 1 as the carrier signal).

[0104] Additionally, it should be noted that although the example above has been described with reference to a configuration of two estimation circuits 308, 309 operating in parallel (implementing two estimation paths), this is not necessary, and the anomaly detection method based on a neural network with an autoencoder architecture, as described above, can also be applied to a motor controller with a single estimation path. Therefore, the safety circuit 313 does not necessarily include a comparator 310 for comparing the outputs of the two estimation paths.

[0105] In summary, according to various embodiments, the following are provided: Figure 12 The motor controller shown in the figure.

[0106] Figure 12 A motor controller 1200 according to an embodiment is shown.

[0107] The motor controller 1200 includes an estimation circuit 1201 configured to perform motor shaft angle estimation using a feedback loop, the feedback loop including feedback of an angle estimation error signal generated by the estimation circuit.

[0108] The motor controller 1200 further includes an anomaly detector 1202, which includes an autoencoder 1203 trained on a segment of the angle estimation error signal.

[0109] Anomaly detector 1202 is configured to supply a segment of the angle estimation error signal generated by estimation circuit 1201 to automatic encoder 1203, determine the reconstruction error of automatic encoder 1203 for the segment of angle estimation error signal, and perform a safety action based on the value derived from the reconstruction error.

[0110] For example, executing such Figure 13The method is illustrated in the figure.

[0111] Figure 13 Flowchart 1300 is shown, illustrating a method for estimating the motor shaft angle and / or the motor shaft rotation speed.

[0112] In 1301, a feedback loop that includes feedback on the angle estimation error signal is used to perform motor shaft angle estimation.

[0113] In 1302, a segment of the angle estimation error signal is supplied to an autoencoder that is trained on the segment of the angle estimation error signal.

[0114] In 1303, the reconstruction error of the automatic encoder for a segment of the angle estimation error signal is determined.

[0115] In 1304, the safety action is performed based on the value derived from the reconstruction error.

[0116] According to various embodiments, in other words, when the angle estimation error signal is encoded and reconstructed (and decoded based on the embedding, i.e., the encoding result), the detection of anomalies (and therefore particularly faults) in the operation of the motor controller (specifically the angle and / or speed estimation part) is performed based on the reconstruction error of the automatic encoder.

[0117] Safety actions could include triggering an alarm or controlling the motor to enter a safety mode, such as stopping it.

[0118] It should be noted that components of the motor controller (e.g., estimation circuitry and anomaly detectors or safety circuitry) can be implemented, and the method can be performed by one or more circuits. In the embodiments, "circuit" can be understood as any kind of logical implementation entity, which can be a dedicated circuit or a processor, firmware, or any combination thereof executing software stored in memory. Therefore, in the embodiments, "circuit" can be hardwired logic circuitry or programmable logic circuitry, such as a programmable processor, e.g., a microprocessor. Thus, "circuit" can be a processor executing software, such as any kind of computer program. According to alternative embodiments, any other kind of implementation of the corresponding functions described herein can also be understood as a "circuit".

[0119] Various examples are described below: Example 1 is as follows (see reference) Figure 12 The motor controller described.

[0120] Example 2 is the motor controller of Example 1, wherein the safety circuit is configured to trigger a safety action as a reaction when the aggregation of reconstruction errors exceeds a predetermined reconstruction error threshold.

[0121] Example 3 is a motor controller of Example 1 or 2, wherein an anomaly detector is configured to, for each of the anomaly detection times in the sequence, supply a segment of the angle estimation error signal generated by the estimation circuit to an automatic encoder, determine the reconstruction error of the automatic encoder for the segment of the angle estimation error signal, and perform a safety action depending on the reconstruction error, wherein the segment of the angle estimation error signal includes the angle estimation error generated by the estimation circuit for the angle estimation time and a predetermined number of angle estimation errors generated by the estimation circuit for the time prior to the angle estimation time.

[0122] Example 4 is the motor controller of Example 3, wherein the anomaly detector is configured to aggregate reconstruction errors determined for multiple anomaly detection times into an aggregate of reconstruction errors, and to perform a safety action depending on the aggregated reconstruction errors.

[0123] Example 5 is a motor controller of Example 4, wherein the safety circuit is configured to trigger a safety action as a reaction when the aggregation of reconstruction error exceeds a predetermined aggregation threshold.

[0124] Example 6 is a motor controller of any one of Examples 1 to 5, wherein an automatic encoder is trained to encode an angle estimation error signal and reconstruct the encoded angle estimation signal.

[0125] Example 7 is a motor controller of any one of Examples 1 to 6, wherein an automatic encoder is trained to encode an angle estimation error signal and reconstruct the encoded angle estimation signal, the angle estimation error signal being generated by the estimation circuit in the absence of a fault.

[0126] Example 8 is a motor controller of any of Examples 1 through 7, where the estimation circuit implements a state observer to estimate the motor shaft angle.

[0127] Example 9 is a motor controller for any of Examples 1 through 8, where the estimation circuit implements a Luenberger observer to estimate the motor shaft angle.

[0128] Example 10 is as follows (refer to) Figure 13 The method described.

[0129] Example 11 is the method of Example 10, which includes training an autoencoder to encode an angle estimation error signal and reconstructing the encoded angle estimation signal.

[0130] Example 12 is a method of Example 10 or 11, which includes training an autoencoder to encode an angle estimation error signal and reconstructing the encoded angle estimation signal, the angle estimation error signal being generated when performing motor shaft angle estimation in the absence of faults.

[0131] It should be noted that the examples and embodiments described in the context of motor controllers are similarly effective for methods used to estimate the motor shaft angle and / or motor shaft rotation speed of a motor, and vice versa.

[0132] Although specific embodiments have been illustrated and described herein, those skilled in the art will appreciate that various alternatives and / or equivalent implementations can be used to replace the illustrated and described embodiments without departing from the scope of the invention. This application is intended to cover any adaptations or variations of the specific embodiments discussed herein. Therefore, it is intended that the invention be limited only to the claims and their equivalents.

[0133] Reference symbol 100 Motor Control Layout 101 Rotary Transformer Rotor 102 Rotary Transformer 103 motor 104 Motor Rotor 105 microcontroller 106 Rotary Transformer-to-Digital Converter 107 Carrier Signal Generator 108 PWM Generator 201 Rotary Transformer-to-Digital Converter 202 Filter Chain Loop 203, 204 ADC 205 Rotary Transformer 206, 207 Mixers 208, 209 Sinc filters 210 Gain Adjuster 211 Carrier Signal Generator 212 State Observer 213 Rotational speed integrator 214 Adders 215 Rotational Angle Integrator 216 Sine LUT 217 Cosine LUT 218 Subtractor 219 Demodulator 220 multipliers 221 Adder 222 Multipliers 300 motor control layout 301 Rotor 302 Rotary Transformer 303 Amplifier 304 Carrier Generator 305 microcontroller 306, 307 ADC 308, 309 Estimation Circuits 310 comparator 311 Anomaly Detector 312 Automatic Encoder 313 Safety Circuit 400 Automatic Encoder 401 Input Layer 402 Hidden Layer 403 Potential Representation 404 Hidden Layer 405 Output Layer 501-503 Signal Diagram 504 Speed ​​Change 505 Peak value of angle estimation error Signal diagrams 601-604 605 Speed ​​Change 701-703 Signal Diagram 801-803 Signal Diagram 901-903 Signal Diagram Signal diagrams 1001-1003 Signal diagrams 1101-1103 1200 Motor Controller 1201 Estimation Circuit 1202 Anomaly Detector 1203 Automatic Encoder 1300 Flowchart Operations 1301-1304

Claims

1. A motor controller, comprising: An estimation circuit is configured to perform motor shaft angle estimation using a feedback loop, the feedback loop including feedback of an angle estimation error signal generated by the estimation circuit; An anomaly detector includes an autoencoder trained on a segment of an angle estimation error signal, wherein the anomaly detector is configured to supply the segment of the angle estimation error signal generated by the estimation circuit to the autoencoder, determine the reconstruction error of the autoencoder for the segment of the angle estimation error signal, and perform a safety action based on the value derived from the reconstruction error.

2. The motor controller of claim 1, wherein the safety circuit is configured to trigger a safety action as a reaction to the aggregation of reconstruction errors exceeding a predetermined reconstruction error threshold.

3. The motor controller according to claim 1 or 2, wherein the anomaly detector is configured to, for each of the anomaly detection times in the sequence, supply a segment of the angle estimation error signal generated by the estimation circuit to the automatic encoder, determine the reconstruction error of the automatic encoder for the segment of the angle estimation error signal, and perform a safety action depending on the reconstruction error, wherein the segment of the angle estimation error signal includes the angle estimation error generated by the estimation circuit for the angle estimation time and a predetermined number of angle estimation errors generated by the estimation circuit for the time prior to the angle estimation time.

4. The motor controller of claim 3, wherein the anomaly detector is configured to aggregate reconstruction errors determined for a plurality of anomaly detection times into an aggregate of reconstruction errors, and to perform a safety action depending on the aggregated reconstruction errors.

5. The motor controller of claim 4, wherein the safety circuit is configured to trigger a safety action as a reaction to the aggregation of reconstruction error exceeding a predetermined aggregation threshold.

6. The motor controller according to any one of claims 1 to 5, wherein the automatic encoder is trained to encode the angle estimation error signal and reconstruct the encoded angle estimation signal.

7. The motor controller according to any one of claims 1 to 6, wherein the automatic encoder is trained to encode an angle estimation error signal and reconstruct the encoded angle estimation signal, the angle estimation error signal being generated by the estimation circuit in the absence of a fault.

8. The motor controller according to any one of claims 1 to 7, wherein the estimation circuit implements a state observer to estimate the motor shaft angle.

9. The motor controller according to any one of claims 1 to 8, wherein the estimation circuit implements a Luenberger observer to estimate the motor shaft angle.

10. A method for estimating the motor shaft angle and / or the motor shaft rotational speed of a motor, comprising: A feedback loop is used to perform motor shaft angle estimation, the feedback loop including feedback of angle estimation error signal; The segment of the angle estimation error signal is fed to the autoencoder that is trained on the segment of the angle estimation error signal; Determine the reconstruction error of the automatic encoder for a segment of the angle estimation error signal; as well as The safety action is performed based on the value derived from the reconstruction error.

11. The method of claim 10, comprising training an autoencoder to encode an angle estimation error signal and reconstructing the encoded angle estimation signal.

12. The method of claim 10 or 11, comprising training an automatic encoder to encode an angle estimation error signal and reconstructing the encoded angle estimation signal, the angle estimation error signal being generated when performing motor shaft angle estimation under fault-free conditions.