Adaptive silent driving system for brushless motor based on foc

Abstract

The application relates to the technical field of brushless motor driving, and discloses a brushless motor adaptive mute driving system based on FOC. The system comprises a state partition module, a strategy mapping module, a parameter combination module, an abnormality updating module and a driving management module. The state partition module collects original electrical signals of the motor, identifies running state characteristics in combination with control parameters, and divides state partitions; the strategy mapping module acquires a pre-stored driving control strategy set, and constructs a mapping relationship between the state partitions and the strategies; the parameter combination module extracts key running parameters, calculates control parameter correlation matching degrees, and generates parameter bias combinations; the abnormality updating module verifies the strategy set in real time, identifies abnormal characteristics, and updates the mapping relationship; and the driving management module identifies target parameters and execution priorities according to the updated mapping relationship, and generates a mute driving monitoring strategy library. The system can adaptively adjust the control strategy, and improves the mute performance of the brushless motor running.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of brushless motor driving technology, in particular to a FOC-based adaptive silent driving system for brushless motor. BACKGROUND

[0002] Brushless motors are widely used in many fields such as household appliances, automobiles, industrial automation, etc. due to their high efficiency and high reliability. With the continuous expansion of application scenarios, the requirement for silent performance during motor operation is becoming increasingly stringent. Currently, the FOC-based brushless motor driving technology has become the mainstream, but in actual operation, it often leads to unstable running state due to factors such as motor load fluctuation, environmental temperature change, manufacturing process difference, etc., thereby generating a large noise.

[0003] Existing driving systems mostly use fixed control strategies, which are difficult to dynamically adjust according to the real-time running state of the motor. For example, when the motor is in a light load state, the fixed control parameters may cause current waveform distortion and increase electromagnetic noise; while in a heavy load state, mechanical vibration noise may be generated due to insufficient torque output. In addition, the noise characteristics of the motor in different speed intervals differ significantly, and the traditional driving system cannot accurately suppress the noise sources in different speed intervals.

[0004] During long-term operation of the motor, its running characteristics gradually change due to reasons such as wear and tear of parts, aging of magnetic circuit, etc., which leads to mismatch between the initially set control strategy and the actual running state, and gradually aggravates the noise problem. The existing driving system lacks real-time monitoring of the motor running state and adaptive updating capability of the control strategy, which is difficult to cope with the dynamic changes during motor operation and cannot fundamentally solve the problem of poor silent effect. SUMMARY

[0005] The purpose of the present application is to provide a FOC-based adaptive silent driving system for brushless motor to solve the problems raised in the background.

[0006] To achieve the above purpose, the present application provides a FOC-based adaptive silent driving system for brushless motor, which comprises:

[0007] A state partition module for collecting original electrical signals during motor operation, identifying running state characteristics according to motor control parameters, and dividing motor running state partitions;

[0008] A strategy mapping module for obtaining a pre-stored driving control strategy set and constructing a mapping relationship between state partitions and driving control strategies;

[0009] a parameter combination module configured to extract key operating parameters from the state partition and the drive control strategy, calculate a correlation matching degree of multiple sets of control parameters in the drive control strategy according to the key operating parameters, and generate a parameter bias combination;

[0010] an abnormality updating module configured to perform real-time verification on the set of drive control strategies based on the parameter bias combination, identify an abnormality feature of the drive control strategy, and update a mapping relationship between the state partition and the drive control strategy according to the abnormality feature;

[0011] a drive management module configured to identify motor operating target parameters and an execution priority according to the updated mapping relationship, and generate a silent drive monitoring strategy library.

[0012] Preferably, the state partition module comprises the following implementation manners:

[0013] a preset state classification model is called for any data frame of the original electrical signal;

[0014] the data frame is marked with a state by using the state classification model, and at least one state category is output;

[0015] current fluctuation features, voltage harmonic features, and torque pulsation features under the corresponding state category are extracted to form an operating feature sample set;

[0016] the current fluctuation features, the voltage harmonic features, and the torque pulsation features in the operating feature sample set are analyzed respectively to generate a current steady-state region, a harmonic analysis region, and a torque pulsation correlation region, which are used as motor operating state partitions.

[0017] Preferably, the strategy mapping module comprises the following implementation manners:

[0018] a preset drive control strategy matched with the state category is called to generate an initial strategy set without an associated state partition label;

[0019] whether the initial strategy set meets a target drive control strategy condition is verified, and the initial strategy set meeting the target condition is marked as a drive control strategy;

[0020] a mapping relationship between the state partition and the drive control strategy is established based on feature descriptions of the current steady-state region, the harmonic analysis region, and the torque pulsation correlation region, and in combination with functional attributes and constraint conditions of the drive control strategy.

[0021] Preferably, the parameter combination module comprises the following implementation manners:

[0022] the state partition and the drive control strategy are cluster analyzed according to working modes, load types, and efficiency thresholds, and a maximum cluster center is selected as a key operating parameter;

[0023] Similarity of core electrical indicators in the calculation of key operating parameters is set, and a public parameter sequence related to the core electrical indicators is set;

[0024] Control parameter groups in the driving control strategy are extracted based on the public parameter sequence, the longest continuous association length between the control parameter groups is set, and the length value is set as the association matching degree;

[0025] According to the association matching degree and the time sequence characteristics of the control parameter groups, a parameter bias combination is generated.

[0026] Preferably, the implementation mode of the abnormality updating module comprises:

[0027] The time sequence characteristics of the control parameter groups are extracted from the parameter bias combination;

[0028] According to the time window corresponding to the time sequence characteristics, a target verification path of the parameter bias combination is divided;

[0029] The target verification path of each control parameter group in the parameter bias combination is fitted, and the probability distribution value of the path in the time window is calculated;

[0030] The probability distribution value is compared with the historical abnormality characteristics of the driving control strategy set, a deviation threshold is marked, and the driving control strategy set is classified and updated according to the deviation threshold, and the mapping relationship between the state partition and the driving control strategy is synchronously updated.

[0031] Preferably, the implementation mode of the driving management module comprises:

[0032] According to the updated mapping relationship, the target operating parameters in the original electrical signal are analyzed;

[0033] According to the appearance frequency of the target operating parameters in the time window, an execution priority is generated;

[0034] The target operating parameters and the execution priority are integrated in a structured data format, and a silent driving monitoring strategy library is output.

[0035] Preferably, the implementation mode of the state partition module further comprises:

[0036] Residual electrical characteristics that do not match the state categories in the operating characteristic sample set are identified;

[0037] The residual electrical characteristics are compared with the standard operating characteristic library, and current difference characteristics, voltage difference characteristics and torque difference characteristics are divided;

[0038] The current fluctuation characteristics that meet the standard operating characteristics in the current steady state region are deleted, and the remaining current difference characteristics are reserved as the current abnormality monitoring region.

[0039] Preferably, the implementation mode of the parameter combination module further comprises:

[0040] extracting a current difference feature, a voltage difference feature and a torque difference feature in the residual electrical characteristics;

[0041] calculating a deviation amount of the current difference feature and a standard current threshold, and constructing a positive deviation sequence and a negative deviation sequence;

[0042] merging continuous deviation amounts in the positive deviation sequence into a first sub-sequence and discrete deviation amounts into a second sub-sequence;

[0043] merging continuous deviation amounts in the negative deviation sequence into a third sub-sequence and discrete deviation amounts into a fourth sub-sequence.

[0044] Preferably, the implementation manner of the anomaly updating module further comprises:

[0045] numerically sorting the deviation amounts in the first sub-sequence and the third sub-sequence;

[0046] taking two adjacent deviation amounts after the sorting as a parameter pair to be verified;

[0047] calling a pre-trained parameter association model to verify whether the parameter pair to be verified meets an electrical association rule;

[0048] counting the number of parameter pairs meeting the electrical association rule.

[0049] Preferably, the implementation manner of the driving management module further comprises:

[0050] calculating a motor operation comprehensive anomaly index according to the number of parameter pairs and the discrete deviation amounts in the second sub-sequence and the fourth sub-sequence;

[0051] when the comprehensive anomaly index exceeds a preset mute threshold, triggering a real-time monitoring instruction and writing a mute driving monitoring strategy library.

[0052] Compared with the prior art, the present application has the following advantages:

[0053] The state partition module collects the original electrical signals during the operation of the motor, and identifies the operation state features in combination with the motor control parameters, thereby achieving accurate division of the motor operation state. This division enables the system to clearly understand the operation characteristics of the motor under different working conditions, thereby providing targeted basis for subsequent control strategy matching.

[0054] The strategy mapping module constructs the mapping relationship between the state partition and the driving control strategy, so that the motor in different operation states can quickly match the corresponding optimal control strategy. Instead of using a single fixed control method, the strategy is flexibly selected according to the real-time operation state of the motor, thereby reducing noise caused by mismatching of the control strategy at the source.

[0055] The parameter combination module extracts key operating parameters from the state partition and the driving control strategy, and calculates the correlation matching degree of multiple sets of control parameters to generate parameter bias combinations. This process further optimizes the combination of control parameters, ensures the synergy between control parameters, and makes the parameters cooperate more reasonably during motor operation, reducing vibration and noise caused by parameter conflicts or incoordination.

[0056] The abnormal update module performs real-time verification on the driving control strategy set based on the parameter bias combination, identifies abnormal characteristics, and updates the mapping relationship. This enables the system to have the ability of self-correction and optimization, and can timely discover and solve the deviation or inapplicability of the control strategy in long-term operation, ensuring that the control strategy is always adapted to the actual operating state of the motor and maintaining good quietness effect.

[0057] The driving management module identifies motor operating target parameters and execution priorities according to the updated mapping relationship, and generates a quiet driving monitoring strategy library. The strategy library provides comprehensive monitoring and control basis for the continuous and stable operation of the motor, can dynamically adjust according to different operating targets and priorities, ensures that the motor can maintain a low noise level under various complex working conditions, and improves the overall quietness performance of the motor operation. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 The timing diagram of the FOC-based brushless motor adaptive quiet driving system described in the present application;

[0059] Figure 2 The flowchart for the state partition module;

[0060] Figure 3 The flowchart for the abnormal update module;

[0061] Figure 4 The flowchart for the residual feature processing of the state partition module;

[0062] Figure 5 The flowchart for parameter verification of the abnormal update module. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0064] Please refer to Figure 1The application provides a FOC-based brushless motor adaptive silent driving system, which comprises an intelligent sensor-based shock absorber body assembly quality inspection system, which comprises a sensing data acquisition module, an assembly state modeling module, a defect analysis module, an experience optimization module, a dynamic adjustment module and a quality inspection result generation module.

[0065] The sensing data acquisition module acquires the original data of vibration sensors, position sensors and pressure sensors in the assembly process in real time, and screens and matches the associated data through a time window. The assembly state modeling module constructs a dynamic model of the shock absorber assembly state based on the sensor data. The defect analysis module extracts position fluctuation and pressure gradient data in the assembly cavity from the model to generate a defect parameter set. The experience optimization module corrects the defect parameters using historical assembly data. The dynamic adjustment module adjusts the path planning and pressure threshold in the quality inspection strategy according to the optimized defect parameters. The quality inspection result generation module finally outputs the assembly operation parameters and generates a quality inspection report.

[0066] Embodiment 1: see Figure 2 In the FOC-based brushless motor adaptive silent driving system, the state partition module is responsible for processing the original electrical signals collected during motor operation. The original electrical signals are acquired in real time through a high-speed ADC converter, including current sampling sequences of each phase of the motor, voltage instantaneous value sequences and feedback signals from the torque sensor. The data frame is transmitted to the internal buffer zone of the module through the system bus, and each data frame contains sampling points within a fixed time interval; for any data frame, the system first performs preprocessing, including noise filtering and data normalization operations. After preprocessing, the module calls the preset state classification model for state recognition. The state classification model is implemented based on a pre-trained convolutional neural network architecture, and the model weights and parameters are stored in the flash memory of the embedded processor. The model input is the feature vector of the preprocessed data frame, including current waveform amplitude, voltage phase angle and torque change rate, etc. The convolutional neural network contains multiple convolutional layers and pooling layers, which are used to extract the spatial features of the signal; the convolution operation acts on the sequence data, and captures the local pattern through the sliding convolution kernel; the activation function adopts ReLU to ensure the nonlinear expression ability, and the final fully connected layer outputs the classification probability. After the model analyzes the data frame, it outputs at least one state category identifier, and the state category is mapped according to the predefined label set, for example, a high load area identifier represents that the motor is running under high torque demand, a low harmonic area identifier indicates that the voltage waveform distortion is small, and a torque stable area identifier indicates that the torque fluctuation is below a set value. The output result is the probability score of one or more categories, and the system selects the output category by setting a threshold and assigns a label.

[0067] After outputting the state category, the module extracts the operating features corresponding to the category. The current fluctuation feature calculation is based on the pre-processed current sample sequence; the module calculates the peak-to-peak difference, i.e., the absolute difference between the maximum and minimum current values, while calculating the average change rate, which is obtained by evaluating the average value of the difference between adjacent sample points. The voltage harmonic feature uses a built-in FFT (Fast Fourier Transform) algorithm for analysis; this algorithm converts the voltage instantaneous value sequence into a frequency domain signal, identifies the fundamental and harmonic components, analyzes the harmonic amplitude distribution, and especially the peak values and proportions of the 3rd, 5th, and 7th harmonic components. The torque pulsation feature is generated based on the standard deviation of the torque sensor feedback data, and the degree of dispersion of the sequence is considered in the standard deviation calculation process, representing the stability of torque change. These feature values are integrated by the feature combiner to form the operating feature sample set; the sample set is organized in a data table format, each sample is associated with a state category label, feature values, and a timestamp, and is stored in a circular buffer to ensure continuous processing without overflow. The feature extraction process uses an interrupt-driven mechanism to avoid CPU resource blocking.

[0068] After generating the operating feature sample set, the module analyzes each feature to divide the motor operating state partition. The analysis process is based on the intrinsic properties of the features: the current fluctuation feature is used to determine the motor operating state; the module evaluates the current peak-to-peak difference and average change rate, and if both the difference and the change rate are below the pre-set static threshold, it is identified as a static operating state, otherwise it is classified as a dynamic operating state. Through this judgment, the module divides the current steady-state region; this region is defined as the operating interval within a certain current fluctuation range, and the boundary parameters include the minimum and maximum fluctuation values, and the output is accompanied by a weight value indicating the confidence of this interval. The analysis of the voltage harmonic feature includes the analysis of the frequency domain data; the module identifies the dominant harmonic frequency range in the FFT result. Based on these frequency ranges, the system forms the harmonic analysis region; this region includes high harmonic and low harmonic subcategories, and the output parameters include the main harmonic frequency list and the related amplitude limit. The analysis of the torque pulsation feature combines the pulse sequence to establish an association mapping; the module calculates the standard deviation of the torque sensor data and compares it with the historical sequence to build a torque and operating state correlation model. The torque pulsation correlation region is thus generated, defining the mapping relationship between the torque change range and the load condition, and the output parameters cover the standard deviation threshold and the correlation probability coefficient. All feature analysis processes use a sliding window mechanism, with a fixed window size of 10 milliseconds of data blocks, each window independently calculates and outputs the partition results, ensuring real-time partition response.

[0069] After partitioning, the module outputs the final motor operating state partition, and the results include three partition objects: current steady-state region, harmonic analysis region, and torque ripple correlation region. Each partition object contains a list of specific parameters, such as the fluctuation boundary parameters of the current steady-state region, the upper and lower frequency limit values of the harmonic analysis region, and the ripple threshold value of the torque ripple correlation region. The partition data is transmitted to the downstream module in a data packet format, including partition labels, timestamps, and feature weight sequences. The system uses a buffering mechanism to manage output, ensuring no data loss under high load. The entire process is executed in a loop on an embedded processor, synchronized with the motor drive frequency, achieving real-time matching with the motor operating state.

[0070] In Example 2, referring to Figure 3 The system performs strategy mapping and parameter combination operations. The strategy mapping module obtains a pre-stored initial set of drive control strategies from the non-volatile memory, including modulation mode parameters, filter coefficient tables, and overcurrent protection threshold configurations. The initial set is not associated with state partition labels and is loaded into the memory buffer in the form of original data tables. The module starts the strategy verification process: each initial strategy is tested for matching with real-time motor operating data, including current waveform, voltage sampling sequence, and torque feedback. The verification criteria are based on the target drive control strategy conditions, which include noise suppression threshold range, efficiency target interval, and torque stability indicators; the output characteristics are monitored in real time during strategy execution, and if the measured noise spectrum value is below the preset limit, the energy conversion rate is higher than the minimum threshold, and the torque fluctuation is within the allowable range, it is determined to meet the target conditions. The initial strategy that meets the verification is marked as the official drive control strategy and added to the mapping construction queue.

[0071] The mapping relationship is established in close combination with the state partition characteristics. The module reads the current steady-state region feature description passed from the upstream, including current fluctuation boundary parameters and confidence coefficients; the harmonic analysis region feature description includes the list of dominant harmonic frequencies and amplitude constraints; the torque ripple correlation region feature description provides the standard deviation threshold and correlation parameters. At the same time, the module analyzes the functional attributes of the drive control strategy: the harmonic suppression strategy corresponds to the high-order harmonic attenuation ability parameter, the efficiency optimization strategy is associated with the power conversion coefficient, and the anti-disturbance strategy is mapped to the torque adjustment response time. The constraints include motor winding temperature limit, maximum allowed current amplitude, and power voltage fluctuation range. The system uses a greedy algorithm to construct the mapping relationship: initialize an empty mapping table, iterate through all state partitions and drive control strategies; for each partition, select the strategy combination that can maximize the satisfaction of functional constraints within the partition feature boundary; each iteration prioritizes strategies with the smallest constraint conflicts to generate a one-to-one correspondence table between partitions and strategies. The mapping table is stored in a hash data structure, with partition identifiers as key values and strategy parameter sets as value objects.

[0072] The parameter combination module receives state partition and drive control strategy data streams. The state partition data contains load type identification (constant / variable load), working mode code (start-up / steady-state / braking) and efficiency threshold parameters; the drive control strategy contains a list of control parameter groups. The module performs multi-level clustering processing: the first layer clusters by working mode, classifying all parameters into start-up cluster, steady-state cluster or braking cluster; the second layer divides sub-clusters within each cluster by load type; the third layer clusters in sub-clusters according to efficiency threshold. The three-dimensional clustering is performed using an optimized K-means algorithm, and the distance calculation is based on the Euclidean distance formula. After clustering is completed, the center point parameters of each cluster are selected as key operating parameters, including typical load change rate, standard efficiency percentage value and reference torque error value. The center point coordinate calculation uses the iterative barycentric adjustment method, and the output is a list of key parameter vectors.

[0073] Based on the key operating parameters, the similarity of core electrical indicators is calculated. The indicators include current harmonic distortion rate sequence, torque response time sequence and energy efficiency conversion rate sequence. For each sequence, a cosine similarity matrix is calculated: the sequence mean vector is obtained, and the dimensional difference is eliminated by standardization processing; the cosine value of the included angle between two sequences is calculated to determine the correlation strength. When setting the common parameter sequence, select the high similarity indicators (cosine value > 0.85) as the correlation object to generate a parameter sequence structure containing the moving average value and standard deviation. The sequence data is stored in a ring buffer in chronological order.

[0074] The module extracts control parameter groups in the drive control strategy, each group containing complete configuration items: SVPWM modulation coefficient vector, PID regulator gain array and dead zone compensation parameter set. The control parameter groups are grouped and matched according to the common parameter sequence: analyze the time feature points in the sequence to locate the active parameter group at the corresponding time; set a time correlation window (default 500ms) and track the parameter group change trajectory within the window. When calculating the continuous correlation length, a dynamic programming algorithm is used to establish a state transition matrix: define the parameter group state space, calculate the feature distance of adjacent time parameter groups; when the feature distance is less than the tolerance threshold, accumulate the continuous count, otherwise reset the count. The longest continuous count is the correlation matching degree value, which is marked in the parameter group metadata. When generating parameter bias combination, the control parameter groups are sorted according to the correlation matching degree priority; high matching degree groups are marked as priority strategies, and low matching degree groups are marked as secondary strategies; the combination contains time sequence labels, parameter preference weights and switching condition parameters, and the data structure is implemented using a priority queue. The final parameter bias combination is output to the abnormal update module through a high-speed data bus.

[0075] Example 3: see Figure 4The abnormal update module receives parameter bias combination data stream from upstream. The combination contains control parameter group sequence and its associated matching degree label. The module first parses the time sequence characteristics of control parameter group: extract the timestamp label of each parameter group, form an ordered time point sequence; at the same time, calculate the derivative characteristics of parameter change, and obtain the change rate data through the numerical difference between the front and rear parameter groups. The time window is divided by a fixed length of 1 second, and the window boundary is synchronized with the system real-time clock, ensuring that each window contains an integer number of control cycles. Based on the time window, the module divides the parameter bias combination into target verification paths: each path corresponds to a parameter group change trajectory within a time window, and the path data includes the starting timestamp, parameter sequence pointer and change rate array. The path segmentation uses a boundary alignment algorithm to avoid data overlap or omission.

[0076] The target verification path fitting process uses a linear regression model. For the control parameter group sequence in each path, the model uses the timestamp as the independent variable and the parameter value as the dependent variable for fitting: establish a least squares optimization problem to solve the linear trend line of parameter change over time. The fitting result generates the slope coefficient and the intercept term, which describes the overall change rule of the parameter in the time window. The probability distribution value calculation uses the Bayesian inference method: based on the fitting result and the historical data distribution, the probability density of the current parameter value appearing in the time window is estimated. The probability density function uses the normal distribution assumption, the mean is determined by the fitting line, and the variance is calculated based on the historical data. The calculation formula of the probability distribution value θ is as follows:

[0077]

[0078] Where: represents the comprehensive probability distribution value in the time window , ; is the standard deviation of the historical parameter value; is the actual parameter value at time t; is determined by the linear fitting function (k is the slope and b is the intercept); the integral interval , corresponds to the start and end time of the time window. The calculation process is implemented by numerical integration, and the discrete sampling point interval is 10ms.

[0079] The historical abnormal feature library is stored in a ring buffer log, containing the execution abnormal records of the driving control strategy in the past 24 hours. The comparison process compares the current theta value with the historical abnormal probability distribution: read the historical theta value distribution of the same time window type, calculate its 0.05 quantile as the lower limit of the deviation threshold, and 0.95 quantile as the upper limit. If the current theta value is lower than the lower limit or higher than the upper limit, it is marked as a deviation threshold event, and a classification symbol containing the deviation direction and amplitude is generated. The classification symbol data structure contains the abnormal type code, the deviation level (1-3 levels) and the timestamp metadata.

[0080] The driving control strategy set update adopts incremental processing: traverse all deviation threshold events, and perform operations on the strategy identifier associated with each event. The low-frequency drift class event triggers the strategy degradation operation: move the corresponding strategy to the standby strategy pool; the high-frequency mutation class event triggers the strategy replacement: select a similar function strategy from the standby pool and insert it into the execution sequence. After the update is completed, the mapping relationship between the state partition and the driving control strategy is reconstructed: the hash mapping table is rebuilt, and the new table dynamically adjusts the key value weight according to the strategy state. The mapping update adopts atomic operation to ensure real-time performance, and the strategy calling interface is frozen during the update.

[0081] The driving management module loads the updated mapping relationship table and parses the target operating parameters in the original electrical signal. The parsing process includes: identifying the silence-related features in the signal through the feature extractor, including the current ripple amplitude minimum value, voltage harmonic distortion peak value and torque fluctuation threshold. The target parameters are stored in a structure format, containing parameter identifier, numerical range and valid period. The time window adopts an overlapping sliding mechanism (window length 1 second, sliding step 0.5 second), and the occurrence frequency of each target parameter in the window is counted: a frequency counter array is maintained, and the corresponding counter is incremented each time the parameter appears.

[0082] The construction of the silent driving monitoring strategy library adopts a hierarchical data structure: the top layer is the time index layer, which is stored in buckets with minute granularity; the middle layer is the parameter layer, which stores the target parameters in each bucket according to priority; the bottom layer is the original data layer, which is associated with the electrical signal sampling point pointer. The output interface is transmitted to the motor control execution unit through the DMA channel, and the transmission rate matches the PWM carrier frequency. The strategy library update period is 500ms, ensuring the control real-time performance.

[0083] Example 4: see Figure 5, the system processes the residual electrical features in the running feature sample set. The state partition module runs a feature differential comparator: traversing all the data frames of the labeled state categories within the sample set, scanning the electrical feature points without assigned category labels; the residual features are extracted to an independent buffer, including current waveform segments, voltage sample outliers and torque pulse outliers. These features are compared with the standard running feature library, which is stored in the off-chip Flash, including the pre-calibrated current baseline waveform, standard harmonic spectrum template and torque expected value curve. The comparison operation uses the absolute value difference algorithm: the current difference feature is the difference between the actual current sample value and the expected current value under the corresponding load in the standard library; the voltage difference feature calculates the Euclidean distance between the actual harmonic component and the standard spectrum; the torque difference feature records the deviation amplitude of the pulse peak value from the standard pulsation interval. The difference values are classified into current / voltage / torque three types of difference feature objects, each object containing a difference type identifier, an absolute difference value and a timestamp.

[0084] The state partition module modifies the existing current steady state region: a feature matching table is established to associate the current fluctuation features with the standard library index, and the matching table uses hash index to speed up the query. Perform matching on each feature point in the current steady state region: if the current fluctuation amplitude and frequency are within the standard fluctuation range, mark it as a matching feature and trigger the deletion instruction; the remaining unmatched points are retained as the current difference feature set. The deletion process uses a lazy recycling mechanism: only the logical deletion bit is marked, and the physical memory is recycled periodically. The retained current difference features are renamed as current anomaly monitoring region, and the region data is attached with a red warning flag bit. The structure includes the maximum deviation, average deviation rate and abnormal duration parameters. The residual feature processing data is shown in Table 1.

[0085] Table 1: Residual feature processing data.

[0086]

[0087] The parameter combination module loads the residual feature data and starts the difference feature serialization process. The current difference features are cached in a sliding window: the window size is 20ms, and the data in the window is sorted by timestamp. Calculate the deviation of each current difference point from the standard current threshold: read the threshold parameter under the current working condition from the EEPROM, and the deviation is calculated as the cumulative algebraic value of the last 3 sampling points. Classify the sequence based on the deviation sign: the positive deviation sequence collects all positive deviation values, and the negative deviation sequence collects all negative deviation values. The sequence elements include the deviation, timestamp and associated state partition code.

[0088] The sub-sequence generation process implements a segmentation merging strategy. In the positive bias sequence, the consecutive bias values are scanned: when the change rate of two adjacent bias values is less than 5%, it is determined as a continuous segment; all bias values in the continuous segment are merged into a first sub-sequence, and the elements are connected by a pointer linked list. The discrete bias values (change rate > 5%) are independently encapsulated as a second sub-sequence. The same rule processes the negative bias sequence: the continuous segment is merged into a third sub-sequence, and the discrete values are merged into a fourth sub-sequence. The sub-sequence data structure includes the start and end time stamps, the average bias value, and the dispersion coefficient. The merging operation uses real-time stream processing: when each new sampling point arrives, the statistics of the latest sub-sequence are updated; when the change rate exceeds the threshold, a new sub-sequence segment is created. During the serialization process, the system maintains the state partition context: when the sub-sequence spans multiple state partitions, it is automatically divided into multiple logical segments, ensuring that each sub-sequence is only associated with a single partition label. The segmentation cutting point is triggered according to the state switching signal, and the cutting process preserves the time stamp continuity. The final output sub-sequence carries metadata, including the associated partition identifier, sequence duration, and maximum bias position.

[0089] In embodiment 5, the anomaly update module receives first sub-sequence and third sub-sequence data from upstream. The first sub-sequence contains a set of positive continuous bias values, and the third sub-sequence contains a set of negative continuous bias values. The module first sorts the bias values in the two sequences independently: using the quicksort algorithm, a monotonically increasing sequence is generated according to the numerical size of the bias values. The sorting process preserves the association between the time stamp and the state partition label. The sequence memory structure is stored in a dynamic array, and after sorting, an ordered bias value queue is formed.

[0090] The sorted sequences are processed in index order: the nth element and the n+1th element in the first sub-sequence form a first parameter pair, and the mth element and the m+1th element in the third sub-sequence form a second parameter pair. The parameter pair selection rule is a skip-1-take-2 mode to avoid overlapping pairing. All parameter pairs are encapsulated into a to-be-verified parameter pair structure, which includes the previous bias value, the next bias value, the time interval value, and the partition identifier. The structure is stored in a double-ended queue buffer area and arranged in ascending order of time stamp.

[0091] The system calls the pre-trained parameter correlation model to perform the verification operation. The model is deployed on the coprocessor and is built based on a support vector machine architecture. An electrical correlation rule library is loaded: the rules define the current-voltage phase difference threshold range, the torque-current linear coefficient boundary, and the power factor tolerance interval. During verification, the model inputs the deviation amount of the parameter pair. The feature vector includes the deviation amount difference, the relative change rate, and the state partition code. The support vector machine kernel function calculates the feature space distance. When the feature vector meets the following conditions, it is determined to comply with the electrical correlation rules: the current deviation change corresponds to the voltage phase difference within the threshold; the torque deviation change rate is linearly related to the current deviation; and the partition identifier matches the power factor constraint range. The verification result is marked in the parameter pair metadata in the form of a Boolean value.

[0092] The statistical mechanism traverses all parameter pairs in the double-ended queue and maintains a counter of parameter pairs that pass verification. The counter has an initial value of zero and is incremented when each parameter pair verification result is true. The counter is stored in a 32-bit hardware register and is automatically cleared and restarted every second to restart the statistical process. The counter value is output in real time to the interrupt event bus of the drive management module.

[0093] The drive management module synchronously receives second and fourth subsequence data: the second subsequence is an array of discrete positive deviations, and the fourth subsequence is an array of discrete negative deviations. The module calculates discrete deviation features: it extracts the absolute deviation amount from the array and calculates the root mean square value as the discrete deviation intensity indicator. In combination with the parameter pair counter value, it performs comprehensive anomaly index calculation. The calculation process includes: dividing the counter value by the total number of parameter pairs per second to obtain the proportion factor R, and the discrete deviation intensity indicator is denoted as D; a weighted fusion formula is used to generate the comprehensive anomaly index . Where α and β are preset weight coefficients stored in the system configuration register.

[0094] After the index is calculated, it is compared with the preset mute threshold. The mute threshold is set to an independent value according to the state partition type: the steady-state region threshold is lower (e.g., 0.7), and the transient-state region threshold is higher (e.g., 1.2). The threshold comparator triggers the judgment at each update: when it exceeds the current partition threshold, it generates a real-time monitoring instruction sequence. The sequence contains three levels of operation instructions: the basic layer increases the current sampling frequency to 200% of the original value; the intermediate layer enables the harmonic spectrum depth diagnosis mode; and the advanced layer activates the torque loop prediction compensation algorithm. The instruction sequence is attached with an effective condition: the basic layer instruction is executed immediately, the intermediate layer instruction is executed until the deviation disappears, and the advanced layer instruction requires manual confirmation.

[0095] The mute drive monitoring strategy library updating process integrates the above outputs. The strategy library creation exception response section data structure section includes: a comprehensive exception index value floating point number, a partition identifier field timestamp millisecond value. The real-time monitoring instruction sequence is encoded as a bit mask: each type of instruction occupies 2 bits, indicating the shutdown standby start state. The data structure is packaged as a binary message frame, written into the strategy library through a dedicated communication interface. The writing period is fixed at 10 ms, covering the 120-139 byte offset address in the strategy library memory area. The interface transmission adopts a redundancy check mechanism to ensure data integrity.

[0096] The system establishes an exception feedback closed loop: when the drive management module writes a new exception response section, it triggers a strategy reload signal to the state partition module. This signal carries the current partition identifier and timestamp parameters, and the drive state classification model is retrained for abnormal period data. The training data extracts historical signal segments from the ring buffer area, and the model weight is updated asynchronously in the isolated memory area. After updating, the model version is automatically switched. Thus, the complete adaptive control closed loop from exception detection to strategy adjustment is completed.

[0097] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0098] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and variations can be made to these embodiments by those skilled in the art without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A brushless motor adaptive silent drive system based on FOC, characterized in that, include: The status partitioning module is used to collect the raw electrical signals of the motor during operation, identify the operating status characteristics based on the motor control parameters, and divide the motor operating status into partitions. The strategy mapping module is used to obtain the pre-stored set of drive control strategies and construct the mapping relationship between state partitions and drive control strategies. The parameter combination module is used to extract key operating parameters from state partitioning and drive control strategy, calculate the correlation matching degree of multiple sets of control parameters in drive control strategy according to key operating parameters, and generate parameter bias combination. The anomaly update module is used to perform real-time verification of the drive control strategy set based on parameter bias combination, identify the abnormal characteristics of the drive control strategy, and update the mapping relationship between the state partition and the drive control strategy according to the abnormal characteristics. The drive management module is used to identify the target parameters and execution priorities of motor operation based on the updated mapping relationship, and generate a silent drive monitoring strategy library. The abnormal update module is implemented in the following ways: Extracting time-series features of control parameter sets from parameter bias combinations; The target verification path is divided based on the time window corresponding to the time series characteristics, with parameter bias combination. The target verification path of each control parameter group in the fitting parameter bias combination is calculated, and the probability distribution value of the path within the time window is calculated. The probability distribution values ​​are compared with the historical abnormal features of the drive control strategy set, deviations from the threshold are marked, and the drive control strategy set is updated according to the deviation threshold. The mapping relationship between the state partition and the drive control strategy is updated synchronously. Identify residual electrical features in the operational feature sample set that do not match the state category; The implementation of the parameter combination module also includes: Extract the current difference feature, voltage difference feature, and torque difference feature from the residual electrical characteristics; Calculate the deviation between the current difference characteristics and the standard current threshold, and construct positive and negative deviation sequences; Continuous deviations in the same state partition of the positive deviation sequence are merged into the first subsequence, and discrete deviations are merged into the second subsequence. The continuous deviations of the same state partition in the negative deviation sequence are merged into the third subsequence, and the discrete deviations are merged into the fourth subsequence. The implementation of the abnormal update module also includes: The deviations in the first and third subsequences are numerically sorted. The two adjacent deviations after sorting are used as the parameter pair to be verified. Call the pre-trained parameter association model to verify whether the parameter pair to be verified satisfies the electrical association rules; Count the number of parameter pairs that conform to the electrical association rules.

2. The FOC-based adaptive silent drive system for brushless motors according to claim 1, characterized in that, The implementation methods of the state partitioning module include: For any data frame of the original electrical signal, invoke the preset state classification model; Use a state classification model to label the state of the data frame and output at least one state category; Extract the current fluctuation characteristics, voltage harmonic characteristics, and torque pulsation characteristics under the corresponding state categories to form an operating characteristic sample set; The current fluctuation characteristics, voltage harmonic characteristics, and torque pulsation characteristics in the operating characteristic sample set are analyzed respectively to generate the current steady-state region, harmonic analysis region, and torque pulsation correlation region, which serve as the motor operating state partitions.

3. The FOC-based adaptive silent drive system for brushless motors according to claim 2, characterized in that, The implementation of the policy mapping module includes: Invoke the preset drive control strategy that matches the state category to generate an initial set of strategies without associated state partition labels; Verify whether the initial policy set meets the target driving control policy conditions, and mark the initial policy set that meets the target conditions as the driving control policy; Based on the characteristic descriptions of the current steady-state region, harmonic analysis region, and torque pulsation correlation region, and combined with the functional attributes and constraints of the drive control strategy, a mapping relationship between state partitions and drive control strategies is established.

4. The FOC-based adaptive silent drive system for brushless motors according to claim 3, characterized in that, The parameter combination module is implemented in the following ways: The state partitioning and drive control strategy are clustered according to working mode, load type and efficiency threshold, and the largest cluster center is selected as the key operating parameter. Calculate the similarity of core electrical indicators among key operating parameters, and set a sequence of common parameters related to the core electrical indicators; Based on the common parameter sequence, the control parameter group in the driving control strategy is extracted, the longest continuous association length between the control parameter groups is set, and this length value is set as the association matching degree. Based on the correlation matching degree and the time series characteristics of the control parameter group, a parameter bias combination is generated.

5. The FOC-based adaptive silent drive system for brushless motors according to claim 4, characterized in that, The driver management module is implemented in the following ways: The target operating parameters in the original electrical signals are analyzed based on the updated mapping relationship; Execution priorities are generated by sorting the target runtime parameters according to their frequency of occurrence within the time window. Integrate target runtime parameters and execution priorities using structured data format, and output a silent drive monitoring strategy library.

6. The FOC-based adaptive silent drive system for brushless motors according to claim 5, characterized in that, The implementation of the state partitioning module also includes: The residual electrical characteristics are compared with the standard operating characteristic library to classify current difference characteristics, voltage difference characteristics, and torque difference characteristics; Remove current fluctuation characteristics that conform to standard operating characteristics in the steady-state current region, and retain the remaining current difference characteristics as the current anomaly monitoring region.

7. The FOC-based adaptive silent drive system for brushless motors according to claim 6, characterized in that, The implementation of the driver management module also includes: The comprehensive abnormal index of motor operation is calculated based on the number of parameter pairs and the discrete deviation in the second and fourth subsequences. When the overall anomaly index exceeds the preset silence threshold, a real-time monitoring command is triggered and written into the silence driver monitoring strategy library.

Images