An electronic controller driving integrated circuit, control method and intelligent terminal

Abstract

The application discloses an electronic controller driving integrated circuit, a control method and an intelligent terminal, relates to the technical field of electronic controller driving control, and comprises the following steps: collecting multi-dimensional original sensing data of a target object including spatial position, time sequence, physical state and environmental parameters through a sensor array; performing synchronous and time sequence calibration processing on the data to generate a fusion data stream with unified space-time reference; performing mode recognition analysis on the fusion data stream, extracting a characteristic behavior mode and calculating a driving demand parameter; generating a driving control sequence including a pulse width modulation waveform instruction and a power stage control parameter based on the parameter; and finally controlling a power switch device to drive a target actuator to act. The method solves the problem of poor fusion quality of multi-source heterogeneous sensing data, and realizes intelligent driving control based on accurate environmental perception and behavior understanding.

Description

Technical Field

[0001] This invention belongs to the field of electronic controller drive control technology, specifically an electronic controller drive integrated circuit, control method and intelligent terminal. Background Technology

[0002] Existing electronic control and drive technologies typically rely on a single type of sensor or a simple combination of a few sensors to acquire the information needed for control. Common approaches involve collecting single physical quantities such as the target object's position and velocity, or independently monitoring state parameters such as temperature and current. Environmental factors are often ignored or handled through independent control loops. This data acquisition lacks a systematic design, is one-dimensional, and has weak correlations between information sources, failing to construct a complete digital mapping of the target object and its surrounding environment.

[0003] Current technologies for processing information from multiple sensors often employ time-division multiplexing, direct transmission of packaged data, or simple weighted summarization. However, due to differences in the operating principles, sampling frequencies, and response times of various sensors, as well as inconsistencies in spatial reference frames caused by different installation locations, the unprocessed data streams inherently exhibit temporal and spatial asynchrony and misalignment. This inconsistency in the underlying data directly leads to inherent contradictions and biases in the information received by subsequent analysis algorithms. Consequently, the reliability of behavior pattern recognition and drive parameter calculations based on such data is insufficient, affecting the accuracy and adaptability of the final control actions. Therefore, acquiring high-quality, spatiotemporally unified fused sensor data streams has become a crucial prerequisite for achieving accurate perception and intelligent actuation. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, the present invention proposes an electronic controller drive control method, comprising: Multi-dimensional raw sensor data of the target object is collected by a sensor array. The multi-dimensional raw sensor data includes spatial location information, time series information, physical state information and environmental parameter information. The multi-dimensional raw sensor data is subjected to data synchronization and timing calibration to generate a fused data stream with a unified spatiotemporal reference. The fused data stream is subjected to pattern recognition analysis to extract the characteristic behavior patterns of the target object and calculate the driving requirement parameters; Based on the drive requirement parameters, a drive control sequence is generated, which includes pulse width modulation waveform instructions and power stage control parameters. The operating state of the power switching device is controlled according to the drive control sequence, and the target actuator is driven to perform an action response.

[0005] Furthermore, the step of performing data synchronization and temporal calibration processing on the multi-dimensional raw sensor data to generate a fused data stream with a unified spatiotemporal reference specifically includes: Establish data cache queues for the spatial location information, time series information, physical state information, and environmental parameter information respectively; Based on the preset master clock source, the time series information in each data buffer queue is re-marked and aligned to the same clock domain; A coordinate transformation matrix is ​​established based on the spatial location information to transform the spatial location information from different sensors to a unified reference coordinate system; Under the unified reference coordinate system and the same clock domain, spatial and temporal interpolation is performed on physical state information and environmental parameter information to fill in data gaps caused by differences in sensor sampling rates; The spatial location information, time series information, physical state information, and environmental parameter information after spatiotemporal calibration and interpolation are encapsulated into a continuous fused data stream in timestamp order.

[0006] Furthermore, the step of performing pattern recognition analysis on the fused data stream, extracting the characteristic behavior patterns of the target object, and calculating the driving requirement parameters specifically includes: The fused data stream is divided into multiple analysis segments according to a time window, and time-domain statistical features and frequency-domain transformation features are extracted from each analysis segment. The time-domain statistical features and frequency-domain transform features are input into a pre-trained classification model, and the classification model outputs a preliminary behavior category corresponding to each analysis segment. Based on the preliminary behavior category, a typical behavior template matching it is retrieved from the historical behavior database. The typical behavior template includes a standard action sequence and its associated energy consumption curve. The physical state information sequence in the current analysis segment is dynamically time-warped and matched with the standard action sequence in the retrieved typical behavior template to calculate the similarity score. When the similarity score exceeds a set threshold, the energy consumption curve associated with the typical behavior template is mapped onto the time axis of the current analysis segment, and the real-time torque and expected speed required for the target actuator to complete the corresponding action are calculated based on the mapping result. The real-time torque and expected speed are the core components of the drive requirement parameters.

[0007] Furthermore, the process of establishing the pre-trained classification model includes: Collect multi-dimensional raw sensor data samples covering various typical behaviors of the target object, and label each data sample with its corresponding real behavior category; The labeled multi-dimensional raw sensor data samples are divided into training set, validation set and test set; Design a deep learning network structure containing convolutional layers and long short-term memory network layers as the backbone network of the classification model; The backbone network is trained using the training set, and the model performance is monitored using the validation set during the training process to prevent overfitting. After training is completed, the recognition accuracy of the classification model is evaluated using the test set, and the qualified model parameters are stored as the pre-trained classification model.

[0008] Furthermore, generating the drive control sequence based on the drive requirement parameters specifically includes: The real-time torque and desired speed are input into a preset motor mathematical model, and the motor mathematical model outputs the target stator current vector required to meet the real-time torque and speed requirements. Based on the target stator current vector, the reference values ​​of the direct-axis current component and quadrature-axis current component in the rotating coordinate system are calculated using a field-oriented control algorithm. The reference values ​​of the direct-axis current component and the quadrature-axis current component are compared with the actual current feedback value obtained through the sampling resistor, and the corresponding control signals of the direct-axis voltage component and the quadrature-axis voltage component are generated by the current regulator. An inverse Parker transformation is performed on the control signals of the direct-axis voltage component and the quadrature-axis voltage component to generate a three-phase voltage modulation wave in a stationary coordinate system; Based on the space vector pulse width modulation strategy, the three-phase voltage modulation wave is compared with the carrier signal to generate six pulse width modulation waveform commands for controlling the power switching device. The six pulse width modulation waveform commands, together with the carrier frequency and dead time parameters, constitute the drive control sequence.

[0009] Furthermore, the adoption of a magnetic field-oriented control algorithm to calculate the reference values ​​of the direct-axis current component and the quadrature-axis current component in the rotating coordinate system specifically includes: The amplitude and phase angle of the target stator current vector are obtained, and the projection component of the target stator current vector in the synchronous rotating coordinate system is calculated in combination with the real-time position electrical angle of the motor rotor. The target stator current vector is decomposed into a direct-axis current component parallel to the rotor magnetic field direction and a quadrature-axis current component perpendicular to the rotor magnetic field direction. Based on the motor electromagnetic torque equation, the real-time torque requirement is converted into the corresponding quadrature axis current component reference value; The reference value of the direct-axis current component is set according to the motor excitation requirements; The reference values ​​of the direct-axis current component and the reference values ​​of the quadrature-axis current component are used as current control reference values ​​in the rotating coordinate system.

[0010] Furthermore, the step of controlling the operating state of the power switching device according to the drive control sequence and driving the target actuator to perform an action response specifically includes: The six pulse width modulation waveform commands are sent to the input port of the gate drive circuit; The gate drive circuit performs level conversion and power amplification on the received waveform command to generate a high-voltage pulse signal with sufficient driving capability. The high-voltage pulse signal is used to control the switching on and off of the power switching device composed of an insulated gate bipolar transistor or a metal-oxide-semiconductor field-effect transistor; The switching action of the power switching device converts the DC bus voltage into a three-phase AC current with the required frequency and voltage amplitude, which is then applied to the motor windings of the target actuator. The three-phase alternating current generates a rotating magnetic field in the motor windings, driving the motor rotor to rotate at the desired speed and output the real-time torque.

[0011] Furthermore, the process of controlling the operating state of the power switching device according to the drive control sequence also includes a closed-loop feedback adjustment step: Real-time sampling of the actual phase current flowing through the power switching device, as well as the actual position and speed of the motor rotor; The actual direct-axis current component and quadrature-axis current component obtained after coordinate transformation of the actual phase current are compared with the reference values ​​of the direct-axis current component and quadrature-axis current component to update the output of the current regulator. The actual position and rotational speed are compared with the desired speed, and the target stator current vector is updated by the speed and position adjuster. The updated target stator current vector is re-input into the motor mathematical model to start a new round of drive control sequence generation loop.

[0012] Furthermore, the present invention also includes an electronic controller driver integrated circuit, the circuit being used to implement the control method as described above, the circuit comprising: The sensor data acquisition module is used to acquire multi-dimensional raw sensor data of the target object through a sensor array. The multi-dimensional raw sensor data includes spatial location information, time series information, physical state information and environmental parameter information. The data fusion processing module is used to perform data synchronization and time-series calibration processing on the multi-dimensional raw sensor data to generate a fused data stream with a unified spatiotemporal reference. The pattern recognition calculation module is used to perform pattern recognition analysis on the fused data stream, extract the characteristic behavior patterns of the target object, and calculate the driving requirement parameters. A control sequence generation module is used to generate a drive control sequence based on the drive requirement parameters, wherein the drive control sequence includes pulse width modulation waveform instructions and power stage control parameters; The power drive output module is used to control the working state of the power switching device according to the drive control sequence, and drive the target actuator to perform an action response.

[0013] Furthermore, the present invention also includes a smart terminal, the terminal comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement an electronic controller drive control method as described above.

[0014] Compared with the prior art, the beneficial effects of the present invention are: By simultaneously collecting four types of information—spatial location, time series, physical state, and environmental parameters—a comprehensive digital description of the target object and its operating scenario is achieved from the data source. This unified and systematic data acquisition scheme transforms the information basis for subsequent analysis and processing from isolated parameter points into a set of interconnected, multi-dimensional states. Pattern recognition algorithms can therefore simultaneously consider multiple factors such as the target's geometric trajectory, its own physical state evolution, and external environmental disturbances, extracting more comprehensive feature behavior patterns and reducing misjudgments caused by limited information dimensions or missing environmental factors.

[0015] Rigorous synchronization and temporal calibration of multi-source heterogeneous raw data generates a fused data stream with a unified intrinsic spatiotemporal reference. This process systematically corrects spatiotemporal misalignments and inconsistencies caused by differences in physical characteristics, sampling frequencies, and installation locations of different sensors, integrating the raw data into highly consistent standardized data entities. Each data unit received by subsequent analysis algorithms precisely corresponds to a complete state snapshot under the same spatiotemporal coordinates, eliminating interference and errors introduced by data asynchrony and coordinate mismatch. This improves the input data quality for feature extraction and behavior pattern recognition, resulting in more reliable and accurate calculated driving requirement parameters.

[0016] The drive control sequence generated based on the aforementioned high-quality fused data and accurately identified behavioral patterns can more realistically map the real-time needs and dynamic changes of the target object through its pulse width modulation waveform commands and power level control parameters. Power switching devices are controlled according to this sequence, driving the actuators to produce action responses. The control logic directly originates from a spatiotemporally precise and dimensionally complete perception and cognition closed loop. The control output of the entire system is no longer based on local or spatiotemporally noisy estimates, but rather on decisions made based on high-fidelity environmental perception and deep behavioral understanding, thereby directly improving the targeting of drive control, dynamic response speed, and overall operational stability. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the electronic controller drive control method described in this invention. Figure 2 The time-domain waveform of the X-direction force component of the six-dimensional force sensor of an industrial robot within a 200ms time window; Figure 3 A flowchart for establishing a pre-trained classification model; Figure 4 A comparison chart of drive control sequence execution and performance; Figure 5 A composite analysis diagram of hierarchical control loops in an industrial electronic controller drive system. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 The electronic controller drive control method acquires multi-dimensional raw sensor data of the target object through a sensor array. The multi-dimensional raw sensor data includes spatial location information, time series information, physical state information, and environmental parameter information. Data synchronization and time series calibration processing is performed on the multi-dimensional raw sensor data to generate a fused data stream with a unified spatiotemporal reference. Pattern recognition analysis is performed on the fused data stream to extract the characteristic behavior patterns of the target object and calculate the drive requirement parameters. Based on the drive requirement parameters, a drive control sequence is generated. The drive control sequence includes pulse width modulation waveform commands and power level control parameters. The working state of the power switching device is controlled according to the drive control sequence to drive the target actuator to perform action response.

[0020] In one embodiment of the present invention, data buffer queues are established for spatial location information, time series information, physical state information, and environmental parameter information, respectively. The time series information in each data buffer queue is re-marked based on a preset master clock source and aligned to the same clock domain. A coordinate transformation matrix is ​​established based on the spatial location information to transform the spatial location information from different sensors to a unified reference coordinate system. Spatial and temporal interpolation is performed on the physical state information and environmental parameter information under the unified reference coordinate system and the same clock domain to fill in the data missing points caused by the difference in sensor sampling rates. The spatial location information, time series information, physical state information, and environmental parameter information after spatiotemporal calibration and interpolation are encapsulated into a continuous fused data stream in timestamp order.

[0021] In specific implementations, the electronic controller-driven control method is applied to the motion control of an industrial robot arm. The sensor array includes a binocular vision sensor, a nine-axis inertial measurement unit (IMU), a six-dimensional force sensor, and a temperature and humidity sensor. The binocular vision sensor acquires the spatial position information of the robot's workpiece gripping point; the nine-axis IMU acquires the time-series information of the robot's joint angular velocity and acceleration; the six-dimensional force sensor acquires the physical state information of the end effector's contact force; and the temperature and humidity sensor acquires the environmental parameter information of the robot's working environment. Data buffer queues are established for the spatial position information, time-series information, physical state information, and environmental parameter information, respectively. The data buffer queues are implemented based on a circular buffer, and each data buffer queue independently stores the corresponding type of raw sensor data. In some embodiments, the time-series information in each data buffer queue is re-marked based on a preset master clock source. The master clock source uses a global clock signal generated by a high-precision crystal oscillator. A unified timestamp is added to the time-series information of the binocular vision sensor, the nine-axis IMU, the six-dimensional force sensor, and the temperature and humidity sensor, respectively. The timestamp accuracy reaches the microsecond level, aligning all time-series information to the same clock domain.

[0022] It is understandable that a coordinate transformation matrix is ​​established based on the spatial location information. This matrix consists of a homogeneous transformation matrix from the camera coordinate system of the binocular vision sensor to the robot's base coordinate system. This transforms the spatial location information from the binocular vision sensor to a unified robot base coordinate system. The transformation process uses the following formula:

[0023] in: This represents the robot's spatial position information in its base coordinate system. This represents the coordinate transformation matrix from the camera coordinate system to the robot's base coordinate system. This represents the spatial position information in the camera coordinate system. In specific implementation, under a unified robot base coordinate system and the same clock domain, spatial and temporal interpolation is performed on the physical state information and environmental parameter information. Spatial interpolation uses a three-dimensional linear interpolation method to supplement the physical state information of points not directly measured by the six-dimensional force sensor within the robot's workspace. Temporal interpolation uses a spline interpolation method to fill in the data gaps caused by the differences in sampling rates between the binocular vision sensor, the nine-axis inertial measurement unit, the six-dimensional force sensor, and the temperature and humidity sensor on the time axis. The sampling rate of the binocular vision sensor is 30 Hz, and the sampling rate of the nine-axis inertial measurement unit is 100 Hz. Through temporal interpolation, a continuous data sequence of all sensors at a sampling rate of 100 Hz is generated.

[0024] Optionally, the spatial location information, time series information, physical state information, and environmental parameter information after spatiotemporal calibration and interpolation are encapsulated into a continuous fused data stream in timestamp order. The data structure of the fused data stream includes timestamp fields, spatial location information fields, time series information fields, physical state information fields, and environmental parameter information fields. Each field stores the calibrated data of the corresponding type. The fused data stream is transmitted to the pattern recognition and analysis module in the form of data packets via a high-speed bus. In some embodiments, the data comparison reflects the consistency of data before and after processing. Before data synchronization and time series calibration, there is a millisecond-level deviation between the timestamp of the spatial location information and the timestamp of the time series information, and there is a rotation and translation deviation between the coordinate system of the spatial location information and the robot base coordinate system. After data synchronization and time series calibration, the spatial location information, time series information, physical state information, and environmental parameter information have a unified time and spatial reference, the time deviation is reduced to within the microsecond level, and all spatial location information is expressed in the robot base coordinate system.

[0025] In one embodiment of the present invention, the fused data stream is divided into multiple analysis segments according to a time window. Time-domain statistical features and frequency-domain transformation features are extracted from each analysis segment. The time-domain statistical features and frequency-domain transformation features are input into a pre-trained classification model. The classification model outputs a preliminary behavior category corresponding to each analysis segment. Based on the preliminary behavior category, a matching typical behavior template is retrieved from the historical behavior database. The typical behavior template includes a standard action sequence and its associated energy consumption curve. The physical state information sequence in the current analysis segment is dynamically time-warped and matched with the standard action sequence in the retrieved typical behavior template. A similarity score is calculated. When the similarity score exceeds a set threshold, the energy consumption curve associated with the typical behavior template is mapped onto the time axis of the current analysis segment. Based on the mapping result, the real-time torque and expected speed required to drive the target actuator to complete the corresponding action are calculated. The real-time torque and expected speed are the core components of the drive requirement parameters.

[0026] In practical implementation, the electronic controller-driven control method is applied to the motion control of an industrial robot arm. The fused data stream contains spatial location information, time series information, physical state information, and environmental parameter information encapsulated in timestamp order. The fused data stream is divided into multiple analysis segments according to a preset time window. The time window of each analysis segment is fixed at 200 milliseconds, and a 50-millisecond overlap region is set between adjacent analysis segments. Time-domain statistical features and frequency-domain transform features are extracted from each analysis segment. The time-domain statistical features calculate the mean, variance, and peak value for the physical state information sequence. The frequency-domain transform features extract the amplitude of the dominant frequency component after performing a fast Fourier transform on the time series information. The time-domain statistical features and frequency-domain transform features are combined into a feature vector and input into a pre-trained classification model. The pre-trained classification model outputs the preliminary behavior category corresponding to each analysis segment. The preliminary behavior categories include "pick up", "rotate", "translate", and "stop". In some embodiments, a typical behavior template matching the preliminary behavior category is retrieved from a historical behavior database. The historical behavior database is stored in an embedded flash memory. The typical behavior template includes a standard action sequence and its associated energy consumption curve. The standard action sequence is a six-dimensional force data sequence with a length of 300 sampling points. The energy consumption curve is a synchronously recorded motor torque and speed curve. The physical state information sequence in the current analysis segment is dynamically time-warped and matched with the standard action sequence in the retrieved typical behavior template. The physical state information sequence comes from the force component sequence in three directions of the six-dimensional force sensor. The dynamic time-warped matching calculates the minimum path cumulative distance between the two sequences as a similarity score.

[0027] In practice, a successful match is determined when the similarity score is below a set threshold of 0.5 Newton-meters. The energy consumption curve associated with the typical behavior template is mapped onto the time axis of the current analysis segment. The mapping process uses linear interpolation based on the time correspondence calculated by dynamic time warping matching to align the time point of the energy consumption curve with the timestamp of the current analysis segment. Based on the mapping results, the real-time torque and expected speed required for the target actuator to complete the corresponding action are calculated. The real-time torque is directly read from the mapped energy consumption curve, and the expected speed is obtained by integrating the rotational speed value from the mapped energy consumption curve. It can be understood that data comparison is reflected in the matching accuracy. When using only a pre-trained classification model for behavior classification, the initial behavior category matches the actual action with 75%. After combining dynamic time warping matching, the sequence similarity assessment between the typical behavior template and the current analysis segment refines the calculation basis of the driving requirement parameters from discrete categories to continuous sequence alignment. The root mean square error of the real-time torque curve is reduced from 15% when directly using the category average curve to 8%. Optionally, the length of the time window is adjusted based on the duration of the target object's behavior. For complex actions lasting more than 500 milliseconds, the time window is extended to 500 milliseconds, and the overlapping area is adjusted to 100 milliseconds to ensure that the entire action is fully covered within one or more analysis segments.

[0028] In some embodiments, after the pre-trained classification model outputs an initial behavior category, the system retrieves all typical behavior templates belonging to that initial behavior category from the historical behavior database. There are ten typical behavior templates. The physical state information sequence in the current analysis segment is then dynamically time-warped and matched with these ten typical behavior templates. The typical behavior template with the lowest similarity score is selected for subsequent mapping. The path search for dynamic time-warping matching uses a dynamic programming algorithm, with the constraint that the path must be continuous and monotonic. It can be understood that during the mapping of the energy consumption curve, the time axis alignment uses a piecewise linear scaling method. Based on the optimal path point pair relationship obtained from dynamic time-warping matching, the time points of the energy consumption curves of the typical behavior templates are mapped to the time points of the current analysis segment. For non-integer mapping points, linear interpolation is used to calculate torque and speed values. Optionally, the calculation of the drive demand parameters includes smoothing of real-time torque and desired speed. The smoothing process uses a first-order low-pass filter with a cutoff frequency of 10 Hz to eliminate high-frequency noise that may be introduced during the mapping process.

[0029] See Figure 2This is a time-domain waveform of the X-direction force component from a six-dimensional force sensor of an industrial robot within a 200ms time window. In the first 0.075 seconds, the force component generally shows an upward trend, with a peak value approaching 4N, corresponding to the stage of contacting the target and applying clamping force during the robot's "pickup" action. Between 0.075 and 0.200 seconds, the force component rapidly decreases to a negative value range (minimum -2N), reflecting the reverse force or inertial impact during action switching. This waveform is a core component of the "fused data stream" in this embodiment. Subsequent steps, such as time window segmentation and FFT frequency domain transformation, will extract feature vectors and input them into a pre-trained model to identify the robot's behavior category. The fluctuation amplitude and frequency of the force component directly reflect the stability of the robot's actions and load changes, serving as the basis for calculating drive requirement parameters.

[0030] In one embodiment of the present invention, see [reference] Figure 3 We collect multi-dimensional raw sensor data samples covering various typical behaviors of the target object, and label each data sample with its corresponding real behavior category. The labeled multi-dimensional raw sensor data samples are divided into training set, validation set and test set. We design a deep learning network structure containing convolutional layers and long short-term memory network layers as the backbone network of the classification model. We use the training set to train the backbone network. During the training process, we use the validation set to monitor the model performance to prevent overfitting. After training, we use the test set to evaluate the recognition accuracy of the classification model, and store the qualified model parameters as a pre-trained classification model.

[0031] In practical implementation, the process of establishing the pre-trained classification model is applied to the behavior recognition of industrial robot arms. Multi-dimensional raw sensor data samples covering various typical behaviors of the target object are collected. These samples originate from joint encoder angle sequences recorded by the robot controller, force and torque sequences recorded by the end effector six-dimensional force sensor, and tool end-point image sequences recorded by the vision sensor. Typical behaviors include "picking up a cylinder," "placing it on a plane," "tightening a screw," and "precision assembly." Each multi-dimensional raw sensor data sample is labeled with its corresponding real behavior category. The labeling is completed synchronously during action execution using a robot programming teach pendant, which records a unique behavior category identifier for each action program. In some embodiments, the labeled multi-dimensional raw sensor data samples are divided into a training set, a validation set, and a test set. The training set contains 70% of the samples in the entire dataset, the validation set contains 15%, and the test set contains 15%. The partitioning process uses stratified random sampling to ensure that the proportion of each behavior category is consistent across the three sets. The partitioned datasets are converted to a unified file format and stored in a solid-state drive array.

[0032] Design a deep learning network structure containing convolutional layers and long short-term memory (LSM) layers as the backbone of a classification model. The convolutional layers are configured as two layers. The first convolutional layer uses 64 sets of size 5 convolutional kernels to perform one-dimensional convolution on the input sensor sequence data and generate a feature map. The second convolutional layer uses 128 sets of size 3 convolutional kernels to perform one-dimensional convolution on the output of the first convolutional layer. The LSM layer is configured as a single layer with 256 memory units and receives the output feature sequence of the convolutional layers as input. The final hidden state of the LSM layer is mapped to the number of behavior categories through a fully connected layer. The fully connected layer is followed by a softmax activation function to output the predicted probability of each category. In practice, the backbone network is trained using a training set. The training process employs mini-batch gradient descent with a batch size of 256. The loss function is classified cross-entropy, and the total number of training epochs is set to 200. The optimizer uses the Adam algorithm with an initial learning rate of 0.001. During training, a validation set is used to monitor model performance to prevent overfitting. The validation set is used to calculate the validation loss after each training epoch. When the validation loss no longer decreases for 10 consecutive training epochs, an early stopping mechanism is triggered to terminate training.

[0033] After training, the classification model's recognition accuracy is evaluated using a test set. The evaluation process involves inputting multi-dimensional raw sensor data samples from the test set into the trained backbone network. The backbone network outputs predicted behavior categories, which are then compared to the true behavior categories to calculate the proportion of correct predictions as the recognition accuracy. The parameters of the evaluated models are then stored as pre-trained classification models. The criterion for passing the evaluation is a recognition accuracy of at least 98.5%. It can be understood that the backbone network training process involves calculating a loss function, the formula for which is:

[0034] in: This represents the average loss function value across all samples in a training batch. This represents the total number of behavior categories. Indicates the sample in category The one-hot encoded value of the real label on the image. This indicates that the sample predicted by the backbone network belongs to a certain category. The probability. Optionally, to prevent the gradient vanishing problem in the early stage of training, a batch normalization layer and a rectified linear unit activation function are introduced after the convolutional layer. The input sequence length of the long short-term memory network layer is fixed at 300 time steps. For samples with a length of less than 300, zero padding is used to the specified length, and for samples with a length of more than 300, truncation is performed.

[0035] In some embodiments, multi-dimensional raw sensor data samples are preprocessed before being input into the backbone network. The preprocessing steps include data normalization and time series alignment. Data normalization linearly scales the data from each sensor channel to zero mean and unit variance. Time series alignment uses a dynamic time warping algorithm to unify the sequence lengths of different samples to a reference length. Optionally, another way to monitor model performance on the validation set to prevent overfitting is to monitor the difference between the validation set accuracy and the training set accuracy. When the difference exceeds 5%, the learning rate is automatically reduced, with a reduction factor set to 0.5. It can be understood that the fixed storage of model parameters includes writing the weight matrices of convolutional layers, the cell state weight matrices of long short-term memory network layers, and the parameter matrices of fully connected layers into the non-volatile memory of the embedded system in binary floating-point format, while simultaneously storing the metadata description file of the network structure.

[0036] In one embodiment of the present invention, the real-time torque and desired speed are input into a preset motor mathematical model. The motor mathematical model outputs a target stator current vector that meets the real-time torque and speed requirements. Based on the target stator current vector, a field-oriented control algorithm is used to calculate the reference values ​​of the corresponding direct-axis current component and quadrature-axis current component in the rotating coordinate system. The reference values ​​of the direct-axis current component and quadrature-axis current component are compared with the actual current feedback value obtained through a sampling resistor. The corresponding control signals of the direct-axis voltage component and quadrature-axis voltage component are generated through a current regulator. An inverse Parker transform is performed on the control signals of the direct-axis voltage component and quadrature-axis voltage component to generate a three-phase voltage modulation wave in the stationary coordinate system. Based on a space vector pulse width modulation strategy, the three-phase voltage modulation wave is compared with a carrier signal to generate a six-channel pulse width modulation waveform command for controlling the power switching device. The six-channel pulse width modulation waveform command, the carrier frequency, and the dead time parameter together constitute the drive control sequence. A field-oriented control algorithm is used to calculate the reference values ​​of the direct-axis current component and quadrature-axis current component in the rotating coordinate system. Specifically, this includes obtaining the amplitude and phase angle of the target stator current vector, calculating the projection component of the target stator current vector in the synchronous rotating coordinate system based on the real-time position electrical angle of the motor rotor, decomposing the target stator current vector into a direct-axis current component parallel to the rotor magnetic field direction and a quadrature-axis current component perpendicular to the rotor magnetic field direction, converting the real-time torque requirement into the corresponding quadrature-axis current component reference value according to the motor electromagnetic torque equation, setting the direct-axis current component reference value according to the motor excitation requirement, and using the direct-axis current component reference value and the quadrature-axis current component reference value as the current control reference value in the rotating coordinate system.

[0037] In practical implementation, the electronic controller drive control method is applied to the joint motor drive control of the industrial robot arm. The real-time torque and the desired speed are input into the preset motor mathematical model. The real-time torque is 2.5 N·m and the desired speed is 120 rpm. The motor mathematical model is the dq axis mathematical model of the permanent magnet synchronous motor. The motor mathematical model outputs the target stator current vector required to meet the real-time torque and speed requirements. The target stator current vector includes an amplitude of 15 amperes and a phase angle of 1.05 radians. In some embodiments, a field-oriented control algorithm is used to calculate the reference values ​​of the direct-axis current component and quadrature-axis current component in the rotating coordinate system based on the target stator current vector. Specifically, this includes obtaining the amplitude and phase angle of the target stator current vector, calculating the projection component of the target stator current vector in the synchronous rotating coordinate system based on the real-time position electrical angle of the motor rotor, decomposing the target stator current vector into a direct-axis current component parallel to the rotor magnetic field direction and a quadrature-axis current component perpendicular to the rotor magnetic field direction, and converting the real-time torque requirement into the corresponding quadrature-axis current component reference value according to the motor electromagnetic torque equation. The motor electromagnetic torque equation is as follows:

[0038] in: This indicates the reference value for the quadrature-axis current component. Indicates real-time torque demand. Indicates the number of pole pairs of the motor. This represents the flux linkage amplitude of the permanent magnet. The reference value of the direct-axis current component is set according to the excitation requirements of the motor. For surface-mounted permanent magnet synchronous motors, the reference value of the direct-axis current component is set to zero amperes to pursue the maximum torque-to-current ratio control. The reference values ​​of the direct-axis current component and the quadrature-axis current component are used as the current control reference values ​​in the rotating coordinate system.

[0039] The reference values ​​of the direct-axis and quadrature-axis current components are compared with the actual current feedback value obtained through a sampling resistor. The actual current feedback value is obtained by sampling the three-phase current through the sampling resistor and performing Clark and Park transforms. The corresponding control signals for the direct-axis and quadrature-axis voltage components are generated by the current regulator. The current regulator is a proportional-integral (PI) regulator with a proportional coefficient of 0.8 and an integral coefficient of 0.05. In specific implementation, the control signals for the direct-axis and quadrature-axis voltage components undergo an inverse Park transform to generate a three-phase voltage modulation wave in a stationary coordinate system. The three-phase voltage modulation wave is a sinusoidal signal with a phase difference of 120 degrees. Based on the space vector pulse width modulation strategy, the three-phase voltage modulation wave is compared with a carrier signal, which is a 10 kHz triangular wave. The comparison process generates six pulse width modulation waveform commands for controlling the power switching devices. The six pulse width modulation waveform commands, along with the carrier frequency and dead time parameters, constitute the drive control sequence. Understandably, the data comparison reflects the response characteristics of the control strategy. Compared to the traditional six-step commutation control, the method of generating drive control sequences based on the motor mathematical model and field-oriented control algorithm reduces the total harmonic distortion rate of the motor phase current from 25% to 6% and the tracking error of the quadrature-axis current component from 15% to 3% under the same real-time torque requirement. Refer to Table 1 for reference values ​​and modulation parameters of the current components under different real-time torque requirements.

[0040] Table 1: Reference values ​​of current components and drive control sequence parameters under different real-time torque requirements Real-time torque demand Expected speed Reference value of quadrature axis current component Direct-axis current component reference value carrier frequency Dead Time 1.0 60 6.0 0.0 10 2.0 2.5 120 15.0 0.0 10 2.0 5.0 180 30.0 0.0 10 2.0 8.0 240 48.0 0.0 10 2.0 Optionally, the motor mathematical model includes the resistance parameters, inductance parameters, and back electromotive force constant of the motor windings. These parameters are obtained from the motor's factory test report and stored in the controller's memory. The space vector pulse width modulation strategy calculates the switching state sequence of the power switching devices by determining the sector and duration of the reference voltage vector. In some embodiments, the output direct-axis voltage component and quadrature-axis voltage component control signals of the current regulator are amplitude-limited. The limiting value is calculated based on the DC bus voltage and motor parameters to prevent over-modulation. It can be understood that the dead time parameter is set according to the turn-on and turn-off time characteristics of the power switching devices to prevent two power switching devices on the same bridge arm from conducting simultaneously. The duty cycle of the six-channel pulse width modulation waveform commands is dynamically determined by the instantaneous comparison result between the three-phase voltage modulation wave and the carrier signal.

[0041] See Figure 4This is a comparison chart of drive control sequence execution and performance, used to evaluate the performance differences between conventional control and field-oriented control (FOC) in terms of total harmonic distortion (THD) and current tracking error under different torque requirements. Across the entire torque range, FOC's THD is only 1 / 3 to 1 / 4 that of conventional control, and its current tracking error is also smaller. This is crucial for scenarios with high precision and efficiency requirements, such as joint motors in industrial robots. As torque demand increases, the THD of both control methods shows a decreasing trend. This is because the motor current amplitude is higher under high torque conditions, diluting the relative harmonic proportion. Figure 4 The study revealed the law that THD decreases with increasing torque, indicating that FOC has a particularly significant advantage under low torque conditions, providing a key selection basis for scenarios such as industrial robot joints that frequently start and stop and operate under low torque.

[0042] In one embodiment of the present invention, a six-channel pulse width modulation waveform command is sent to the input port of the gate drive circuit. The gate drive circuit performs level conversion and power amplification on the received waveform command to generate a high-voltage pulse signal with sufficient driving capability. The high-voltage pulse signal is used to control the switching on and off of a power switching device composed of an insulated gate bipolar transistor or a metal-oxide-semiconductor field-effect transistor. The switching action of the power switching device converts the DC bus voltage into a three-phase alternating current with the required frequency and voltage amplitude, which is applied to the motor winding of the target actuator. The three-phase alternating current generates a rotating magnetic field in the motor winding, driving the motor rotor to rotate at the desired speed and output real-time torque. The process of controlling the working state of the power switching device according to the drive control sequence also includes a closed-loop feedback adjustment step. The actual phase current flowing through the power switching device and the actual position and speed of the motor rotor are sampled in real time. The actual direct-axis current component obtained after coordinate transformation of the actual phase current is compared with the reference values ​​of the quadrature-axis current component and the direct-axis current component and the quadrature-axis current component to update the output of the current regulator. The actual position and speed are compared with the desired speed, and the target stator current vector is updated through the speed and position regulator. The updated target stator current vector is re-input into the motor mathematical model to start a new cycle of drive control sequence generation.

[0043] In practical implementation, the electronic controller drive control method is applied to the permanent magnet synchronous motor drive of the joint of the industrial robot arm. Six pulse width modulation waveform commands are sent to the input port of the gate drive circuit. The six pulse width modulation waveform commands come from the output of the space vector pulse width modulation strategy. The waveform commands are logic level signals. The gate drive circuit performs level conversion and power amplification on the received waveform commands to generate a high voltage pulse signal with sufficient driving capability. The amplitude of the high voltage pulse signal is 15 volts and the rise time is less than 100 nanoseconds. In some embodiments, a high-voltage pulse signal is used to control the operating state of a power switching device composed of an insulated-gate bipolar transistor (IGBT). The power switching device is arranged in a three-phase full-bridge configuration. The high-voltage pulse signal is directly applied to the gate pin of the IGBT to control its on / off state. The collector and emitter of the IGBT are subjected to a DC bus voltage of 310 volts. The switching action of the power switching device converts the DC bus voltage into a three-phase alternating current with the required frequency and voltage amplitude. The required frequency is calculated to be 40 Hz based on the desired speed, and the voltage amplitude is the line voltage of 110 volts. This is applied to the motor windings of the target actuator. The motor windings are three-phase stator windings connected in a star configuration. The three-phase alternating current generates a rotating magnetic field in the three-phase stator windings. The rotating magnetic field drives the motor rotor to rotate at the desired speed and outputs real-time torque.

[0044] The process of controlling the operating state of the power switching devices according to the drive control sequence also includes a closed-loop feedback regulation step. This involves real-time sampling of the actual phase current flowing through the power switching devices, as well as the actual position and speed of the motor rotor. The actual phase current is obtained through a sampling resistor connected in series with the lower transistor of the three-phase bridge arm, with a resistance of 0.01 ohms. The actual position and speed of the motor rotor are obtained through an optical encoder mounted on the motor shaft, which outputs 2,500 pulses per revolution. It can be understood that the actual direct-axis current component and quadrature-axis current component obtained after coordinate transformation are compared with reference values ​​for the direct-axis and quadrature-axis current components. The coordinate transformation includes Clarke transform and Park transform. The comparison process is completed in the interrupt service routine of the digital signal processor, updating the output of the current regulator. The output of the current regulator is the updated control signal for the direct-axis voltage component and quadrature-axis voltage component. In specific implementation, the actual position and speed are compared with the desired speed. This comparison process is completed in the main loop of the digital signal processor. The target stator current vector is updated through the speed and position regulator, which uses a proportional-integral (PI) regulation algorithm. Its control law is described by the following formula:

[0045] in: This represents the magnitude update of the target stator current vector. This represents the speed loop proportionality coefficient. Indicates the desired speed. Indicates the integral coefficient of the velocity loop. This indicates the actual rotational speed of the motor rotor. This represents the integral term of the speed error. The updated target stator current vector is re-input into the motor mathematical model, initiating a new cycle of drive control sequence generation.

[0046] Data comparison is reflected in the response of the control loop. When only open-loop control is executed, the actual speed of the motor will overshoot by 20% under a step change in load, and the recovery time is 500 milliseconds. After introducing the closed-loop feedback regulation step, the overshoot of the actual speed is reduced to less than 5%, and the recovery time is shortened to 100 milliseconds. Optionally, the execution frequency of the closed-loop feedback regulation step is set in layers according to control requirements. The regulation frequency of the current loop is set to 10 kHz, and the regulation frequency of the speed and position loop is set to 1 kHz. The gate drive circuit integrates short-circuit protection and undervoltage lockout functions to prevent damage to the power switching devices. In some embodiments, the target stator current vector amplitude update value output by the speed and position regulator is limited. The limiting value is set according to the rated current of the motor and the maximum output capability of the driver. The phase angle of the updated target stator current vector is calculated based on the actual position electrical angle of the motor rotor and the desired power factor angle. It can be understood that the generation cycle of the new round of drive control sequence is synchronized with the current switching state of the power switching devices. When the digital signal processor generates a new six-channel pulse width modulation waveform instruction, it will consider the current state of the power switching devices to ensure a smooth transition of switching action and avoid voltage spikes.

[0047] See Figure 5 This is a composite analysis diagram of the hierarchical control loops of an industrial electronic controller drive system. The diagram clearly illustrates the hierarchical control logic of the drive system. The fault protection layer responds the fastest, followed by the current loop and PWM modulation layer, with the speed and position loop responding the slowest. This perfectly matches the typical design of an industrial permanent magnet synchronous motor drive. The differences in error rates reflect the accuracy requirements of different loops. The protection and PWM modulation layers have the highest requirements for real-time performance and accuracy, while the speed loop focuses more on the smoothness of dynamic response. The control quality of each loop can be directly evaluated through the quantified error rate values. For example, a 0.2% error rate in the current loop demonstrates its d / q axis current tracking capability, while a 0.1% error rate in PWM modulation reflects the accuracy of waveform generation.

[0048] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. An electronic controller drive control method, characterized in that, The method includes: Multi-dimensional raw sensor data of the target object is collected by a sensor array. The multi-dimensional raw sensor data includes spatial location information, time series information, physical state information and environmental parameter information. The multi-dimensional raw sensor data is subjected to data synchronization and timing calibration to generate a fused data stream with a unified spatiotemporal reference. The fused data stream is subjected to pattern recognition analysis to extract the characteristic behavior patterns of the target object and calculate the driving requirement parameters; Based on the drive requirement parameters, a drive control sequence is generated, which includes pulse width modulation waveform instructions and power stage control parameters. The operating state of the power switching device is controlled according to the drive control sequence, and the target actuator is driven to perform an action response.

2. The electronic controller drive control method according to claim 1, characterized in that, The step of performing data synchronization and temporal calibration on the multi-dimensional raw sensor data to generate a fused data stream with a unified spatiotemporal reference specifically includes: Establish data cache queues for the spatial location information, time series information, physical state information, and environmental parameter information respectively; Based on the preset master clock source, the time series information in each data buffer queue is re-marked and aligned to the same clock domain; A coordinate transformation matrix is ​​established based on the spatial location information to transform the spatial location information from different sensors to a unified reference coordinate system; Under the unified reference coordinate system and the same clock domain, spatial and temporal interpolation is performed on physical state information and environmental parameter information to fill in data gaps caused by differences in sensor sampling rates; The spatial location information, time series information, physical state information, and environmental parameter information after spatiotemporal calibration and interpolation are encapsulated into a continuous fused data stream in timestamp order.

3. The electronic controller drive control method according to claim 2, characterized in that, The process of performing pattern recognition analysis on the fused data stream, extracting the characteristic behavior patterns of the target object, and calculating the driving requirement parameters specifically includes: The fused data stream is divided into multiple analysis segments according to a time window, and time-domain statistical features and frequency-domain transformation features are extracted from each analysis segment. The time-domain statistical features and frequency-domain transform features are input into a pre-trained classification model, and the classification model outputs a preliminary behavior category corresponding to each analysis segment. Based on the preliminary behavior category, a typical behavior template matching it is retrieved from the historical behavior database. The typical behavior template includes a standard action sequence and its associated energy consumption curve. The physical state information sequence in the current analysis segment is dynamically time-warped and matched with the standard action sequence in the retrieved typical behavior template to calculate the similarity score. When the similarity score exceeds a set threshold, the energy consumption curve associated with the typical behavior template is mapped onto the time axis of the current analysis segment, and the real-time torque and expected speed required for the target actuator to complete the corresponding action are calculated based on the mapping result. The real-time torque and expected speed are the core components of the drive requirement parameters.

4. The electronic controller drive control method according to claim 3, characterized in that, The process of building the pre-trained classification model includes: Collect multi-dimensional raw sensor data samples covering various typical behaviors of the target object, and label each data sample with its corresponding real behavior category; The labeled multi-dimensional raw sensor data samples are divided into training set, validation set and test set; Design a deep learning network structure containing convolutional layers and long short-term memory network layers as the backbone network of the classification model; The backbone network is trained using the training set, and the model performance is monitored using the validation set during the training process to prevent overfitting. After training is completed, the recognition accuracy of the classification model is evaluated using the test set, and the qualified model parameters are stored as the pre-trained classification model.

5. The electronic controller drive control method according to claim 3, characterized in that, The step of generating a drive control sequence based on the drive requirement parameters specifically includes: The real-time torque and desired speed are input into a preset motor mathematical model, and the motor mathematical model outputs the target stator current vector required to meet the real-time torque and speed requirements. Based on the target stator current vector, the reference values ​​of the direct-axis current component and quadrature-axis current component in the rotating coordinate system are calculated using a field-oriented control algorithm. The reference values ​​of the direct-axis current component and the quadrature-axis current component are compared with the actual current feedback value obtained through the sampling resistor, and the corresponding control signals of the direct-axis voltage component and the quadrature-axis voltage component are generated by the current regulator. An inverse Parker transformation is performed on the control signals of the direct-axis voltage component and the quadrature-axis voltage component to generate a three-phase voltage modulation wave in a stationary coordinate system; Based on the space vector pulse width modulation strategy, the three-phase voltage modulation wave is compared with the carrier signal to generate six pulse width modulation waveform commands for controlling the power switching device. The six pulse width modulation waveform commands, together with the carrier frequency and dead time parameters, constitute the drive control sequence.

6. The electronic controller drive control method according to claim 5, characterized in that, The aforementioned magnetic field-oriented control algorithm calculates reference values ​​for the direct-axis and quadrature-axis current components in the rotating coordinate system, specifically including: The amplitude and phase angle of the target stator current vector are obtained, and the projection component of the target stator current vector in the synchronous rotating coordinate system is calculated in combination with the real-time position electrical angle of the motor rotor. The target stator current vector is decomposed into a direct-axis current component parallel to the rotor magnetic field direction and a quadrature-axis current component perpendicular to the rotor magnetic field direction. Based on the motor electromagnetic torque equation, the real-time torque requirement is converted into the corresponding quadrature axis current component reference value; The reference value of the direct-axis current component is set according to the motor excitation requirements; The reference values ​​of the direct-axis current component and the reference values ​​of the quadrature-axis current component are used as current control reference values ​​in the rotating coordinate system.

7. The electronic controller drive control method according to claim 5, characterized in that, The step of controlling the operating state of the power switching device according to the drive control sequence and driving the target actuator to perform an action response specifically includes: The six pulse width modulation waveform commands are sent to the input port of the gate drive circuit; The gate drive circuit performs level conversion and power amplification on the received waveform command to generate a high-voltage pulse signal with sufficient driving capability. The high-voltage pulse signal is used to control the switching on and off of the power switching device composed of an insulated gate bipolar transistor or a metal-oxide-semiconductor field-effect transistor; The switching action of the power switching device converts the DC bus voltage into a three-phase AC current with the required frequency and voltage amplitude, which is then applied to the motor windings of the target actuator. The three-phase alternating current generates a rotating magnetic field in the motor windings, driving the motor rotor to rotate at the desired speed and output the real-time torque.

8. The electronic controller drive control method according to claim 7, characterized in that, The process of controlling the operating state of the power switching device according to the drive control sequence also includes a closed-loop feedback adjustment step: Real-time sampling of the actual phase current flowing through the power switching device, as well as the actual position and speed of the motor rotor; The actual direct-axis current component and quadrature-axis current component obtained after coordinate transformation of the actual phase current are compared with the reference values ​​of the direct-axis current component and quadrature-axis current component to update the output of the current regulator. The actual position and rotational speed are compared with the desired speed, and the target stator current vector is updated by the speed and position adjuster. The updated target stator current vector is re-input into the motor mathematical model to start a new round of drive control sequence generation loop.

9. An electronic controller driver integrated circuit, characterized in that, The circuit is used to implement the control method according to any one of claims 1 to 8, the circuit comprising: The sensor data acquisition module is used to acquire multi-dimensional raw sensor data of the target object through a sensor array. The multi-dimensional raw sensor data includes spatial location information, time series information, physical state information and environmental parameter information. The data fusion processing module is used to perform data synchronization and time-series calibration processing on the multi-dimensional raw sensor data to generate a fused data stream with a unified spatiotemporal reference. The pattern recognition calculation module is used to perform pattern recognition analysis on the fused data stream, extract the characteristic behavior patterns of the target object, and calculate the driving requirement parameters. A control sequence generation module is used to generate a drive control sequence based on the drive requirement parameters, wherein the drive control sequence includes pulse width modulation waveform instructions and power stage control parameters; The power drive output module is used to control the working state of the power switching device according to the drive control sequence, and drive the target actuator to perform an action response.

10. A smart terminal, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements an electronic controller drive control method as described in any one of claims 1 to 8.

Images