Methods for compensating for gear backlash and tooth force in reducers of new energy vehicles
By dynamically calculating the real-time optimal gear force using an AI self-learning model, the problems of transmission accuracy and noise in the gear backlash compensation method of new energy vehicle reducers are solved, adaptive compensation is achieved, and the accuracy and reliability of the transmission system are improved.
Patent Information
- Application Number
- CN202611108600.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-24
- Publication Date
- 2026-11-10
AI Technical Summary
In the existing technology, the compensation method for gear backlash in new energy vehicle reducers cannot adapt to dynamic changes caused by factors such as gear wear, temperature, and load, resulting in reduced transmission accuracy and increased noise. Furthermore, it relies on manual experience, which leads to low efficiency and poor consistency.
By employing an AI self-learning model combined with multi-dimensional factor data collection and using a CNN-LSTM hybrid neural network architecture, the system dynamically calculates the real-time optimal gear force to achieve adaptive compensation for gear backlash in the reducer. This includes real-time adjustments to gear backlash baseline data, motor torque, vibration, temperature, and mileage data.
It significantly improves transmission accuracy and smoothness, reduces noise, reduces reliance on human experience, extends reducer life, and lowers production and maintenance costs.
Smart Images

Figure CN123067104A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of automotive technology. Specifically, this invention relates to a method for compensating for gear backlash and tooth force in a reducer of a new energy vehicle. Background Technology
[0002] The reducer of a new energy electric vehicle needs to convert the high speed of the drive motor into the low speed required by the wheels, and its gear meshing surface needs to have a small gap to ensure flexibility.
[0003] Existing technologies generally employ fixed-value compensation to counteract the lag and vibration caused by the gear clearance during motor reversal. However, gear clearance is dynamically affected by factors such as wear, temperature, and load, and exhibits a non-linear distribution across the entire speed range. This leads to a deterioration in the effectiveness of fixed compensation values after long-term operation, making them unsuitable for different operating conditions.
[0004] In addition, clearance calibration relies on manual experience, which is inefficient, inconsistent, and prone to zero-crossing noise; it also lacks self-adaptive ability, and long-mileage wear will deteriorate transmission smoothness and affect gear life.
[0005] This invention provides a method for compensating for gear backlash and tooth force in a reducer of a new energy vehicle, particularly for improving the transmission accuracy of an electric drive system in a new energy vehicle. Summary of the Invention
[0006] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention provides a method for compensating for gear backlash and tooth force in a reducer of a new energy vehicle, with the purpose of improving the transmission accuracy of the electric drive system of a new energy vehicle.
[0007] To achieve the above objectives, the technical solution adopted by this invention is: a method for compensating gear backlash and tooth force in a reducer of a new energy vehicle, used to compensate and control the gear backlash of the reducer in the electric drive system of a new energy vehicle, comprising the following steps:
[0008] S1. Multi-dimensional factor data acquisition: Acquire multi-dimensional influencing factor data related to the gear clearance of the reducer in new energy vehicles. The multi-dimensional influencing factor data includes gear clearance reference data, motor torque data, reducer vibration reference data, temperature correlation coefficient data, mileage data, and gear force reference data.
[0009] S2. AI self-learning model processing: Input the multi-dimensional influencing factor data obtained in step S1 into the pre-trained AI self-learning model, and output the weight coefficients corresponding to each influencing factor through the AI self-learning model.
[0010] S3. Adaptive tooth-aligning force calculation: Based on the weight coefficients output in step S2 and the multi-dimensional influencing factor data obtained in step S1, an adaptive tooth-aligning force calculation formula is established, and the real-time optimal tooth-aligning force under the current working condition is obtained through the calculation formula.
[0011] S4. Dynamic compensation execution: Based on the real-time optimal gear force obtained in step S3, a command is sent to the motor controller to control the motor to output the corresponding torque, so that the meshing surface of the reducer gears remains in contact, thereby realizing real-time compensation of the reducer gear backlash.
[0012] S5. Model Iteration and Optimization: Continuously collect the transmission state data of the reducer after gear backlash compensation, and feed the transmission state data back to the AI self-learning model for iterative training, dynamically updating the weight coefficients corresponding to each influencing factor.
[0013] In step S1, the gear backlash reference data is obtained in the following manner:
[0014] During the reducer bench test and the vehicle EOL off-line testing, the gear backlash self-learning command is initiated by the motor controller to control the motor output torque within ±5 N.m. The electrical angle change value of the drive motor during the process of the reducer gear switching from positive tooth surface contact to negative tooth surface contact is recorded. The average value of the electrical angle change value collected N times is taken to obtain the gear backlash reference value △θ.
[0015] The gear clearance reference value △θ is correlated and corrected with the mileage data. The specific correction method is as follows:
[0016] Using 5000km as the step size, when the cumulative mileage of the vehicle reaches the preset number of kilometers, under the condition that the vehicle is powered down for the first time and the safe driving conditions are met, the gear backlash self-learning command is initiated through the motor controller to control the drive motor to output a torque of ±2N.m, record the change value of the electric angle of the drive motor during the process of the reducer gear switching from positive tooth surface contact to negative tooth surface contact, and correct the gear backlash reference value △θ based on the change value of the electric angle.
[0017] In step S1, the motor torque data is the torque command value T of the motor controller. The torque command value T is read in real time through the vehicle controller via the CAN bus, with a sampling frequency of not less than 100Hz. The torque command value T covers the torque signal of the entire vehicle under all driving, braking, and coasting conditions. After the acquisition is completed, the torque signal is filtered and denoised.
[0018] In step S1, the vibration reference data of the reducer is obtained in the following manner:
[0019] During the end-of-life (EOL) testing phase of the vehicle, vibration data of the reducer in the 50Hz-8000Hz frequency band is collected by vibration sensors installed at preset key positions on the reducer housing. The root mean square value of the vibration data is calculated to obtain the reducer vibration reference value G.
[0020] In step S1, the temperature correlation coefficient data is the linear relationship coefficient Kt between the reducer oil temperature and ambient temperature, motor torque, and motor speed. The linear relationship coefficient Kt is calculated in the following way:
[0021] Beforehand, the fitting constants d, e, and f are calibrated through a steady-state thermal test on the reducer bench, and a linear correlation model between oil temperature and ambient temperature, torque, and speed is established. During the real-time operation of the vehicle, real-time motor torque T, motor speed n, ambient temperature Tenv, and real-time reducer oil temperature Toil are collected, and the linear relationship coefficient Kt is calculated using the formula Kt=(Toil-Tenv) / (d*T+e*n+f).
[0022] In step S1, the mileage data is the vehicle's cumulative mileage M, which is read in real time from the vehicle's instrument system with an accuracy of not less than 1km; the weighting coefficient k5 corresponding to the vehicle's cumulative mileage M is dynamically adjusted according to the mileage interval.
[0023] When M≤5000km, the value of k5 ranges from 0.008 to 0.02;
[0024] When 5000km < M ≤ 80000km, the value of k5 ranges from 0.02 to 0.07.
[0025] When M > 80000 km, the value of k5 ranges from 0.07 to 0.11.
[0026] In step S1, the tooth-mounting force reference data is the tooth-mounting force reference value F, which is obtained through calibration via actual vehicle road testing. The specific calibration method is as follows:
[0027] The tooth force output of the drive motor is adjusted under various working conditions, including urban roads, highways, and mountain roads. The output range of the tooth force is ±5N, the control accuracy is no more than ±0.3N, and the response time is no more than 30ms. Combining subjective driving evaluation with objective test data of impact and vibration, the tooth force value corresponding to the optimal transmission smoothness of the reducer is determined as the tooth force reference value F.
[0028] The AI self-learning model adopts a CNN-LSTM hybrid neural network architecture. The input layer of the AI self-learning model standardizes the input multi-dimensional influencing factor data. The hidden layer includes two convolutional layers and three LSTM layers. The output layer outputs the weight coefficients k0-k5 corresponding to each influencing factor. The training convergence error of the AI self-learning model does not exceed 2%. The AI self-learning model is pre-trained with the optimal smoothness of the reducer transmission as the objective function, and learns the nonlinear mapping relationship between multi-dimensional influencing factor data and gear force.
[0029] The adaptive tooth-aligning force calculation formula is as follows:
[0030]
[0031] Wherein, Fk is the real-time optimal gear force, F is the gear force reference value, △θ is the gear backlash reference value, T is the torque command value of the motor controller MCU, G is the reducer vibration reference value, Kt is the temperature correlation coefficient, M is the vehicle's cumulative mileage, k0 is the reference compensation coefficient, and k1-k5 are the weight coefficients corresponding to each influencing factor output by the AI self-learning model.
[0032] In step S5, the reducer transmission state data includes compensated reducer vibration data, drive motor torque data, drive motor position data, and reducer oil temperature data. The AI self-learning model continuously iterates based on the transmission state data to continuously adapt the compensation accuracy to changes in the vehicle's operating conditions and the wear state of the reducer gears.
[0033] The method for compensating for gear backlash and tooth force in a new energy vehicle reducer according to the present invention has the following beneficial effects:
[0034] 1. High compensation accuracy: By integrating six core factors, the AI model characterizes nonlinear mapping relationships, with a compensation error of ≤2%, which is significantly better than traditional methods;
[0035] 2. Strong adaptive robustness: It dynamically adapts to changes in operating conditions such as temperature, torque, and mileage. It combines offline calibration learning with periodic self-learning prediction to dynamically update clearance data and compensation strategies, adapt to different operating conditions, extend the service life of the reducer, and eliminate the need for manual parameter adjustment throughout its entire life cycle.
[0036] 3. Superior transmission performance: Effectively reduces gear impact noise, improves transmission smoothness by more than 40%, and provides a better driving experience;
[0037] 4. Reduce costs and reliance: Reduce reliance on manual experience and high-precision hardware processing, improve production and debugging efficiency, and reduce maintenance costs and the difficulty of large-scale production. Attached Figure Description
[0038] This manual includes the following figures, which illustrate the following:
[0039] Figure 1 This is a flowchart of the method for compensating for gear backlash and tooth force in a new energy vehicle reducer according to the present invention. Detailed Implementation
[0040] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, in order to help those skilled in the art to have a more complete, accurate and in-depth understanding of the concept and technical solutions of the present invention, and to facilitate its implementation.
[0041] It should be noted that in the following embodiments, the terms "first," "second," and "third" do not represent an absolute distinction in structure and / or function, nor do they represent the order of execution; they are merely for the convenience of description.
[0042] The technical concept of this invention includes: a reducer for new energy vehicles, a core transmission component of the electric drive system of new energy electric vehicles, responsible for converting the high speed and low torque of the drive motor into the low speed and high torque required by the wheels. Its transmission accuracy directly affects the vehicle's driving smoothness, power response speed, and energy efficiency. To ensure transmission flexibility and avoid low-temperature jamming, a small gap (i.e., gear clearance) needs to be reserved on the gear meshing surface. However, this gap will cause transmission lag, positioning errors, and vibration noise when the motor drives in reverse, especially under frequent reversing conditions such as rapid acceleration, rapid deceleration, and reversing, the problem is more prominent.
[0043] In existing technologies, new energy electric vehicles generally adopt a fixed-value compensation method, which involves preset a fixed reverse clearance value and instructing the motor to rotate an additional angle to compensate for the clearance when it reverses. However, gear clearance exhibits significant time-varying characteristics due to factors such as mileage (wear), temperature load (thermal expansion and contraction), and road conditions (impact load). Fixed compensation values cannot be matched in real time, leading to a deterioration in compensation effect after long-term operation. Furthermore, the clearance exhibits a non-linear distribution across the entire speed range, making it difficult for fixed values to accurately cover the needs of different operating conditions such as urban congestion, high-speed cruising, and hill driving. Although some studies have attempted to optimize the compensation effect through complex physical modeling, this requires extensive vehicle road testing and debugging, resulting in poor adaptability and inability to meet the dynamic control requirements of new energy electric vehicles under all operating conditions. The shortcomings of existing technologies include: 1. The calibration and adjustment of reducer gear clearance rely on manual experience, leading to low efficiency and poor consistency, resulting in zero-crossing knocking noises in different vehicles; 2. Gear clearance compensation methods are mostly based on fixed parameters, failing to adapt to dynamic changes in clearance caused by gear wear, mileage, load, and speed variations, lacking adaptive capability. 3. Long-distance gear wear and impacts can lead to poor transmission smoothness and affect gear durability. The technical solution of this invention is as follows:
[0044] like Figure 1 As shown, this embodiment of the invention provides a method for compensating gear backlash and gear force in a new energy vehicle reducer, used to compensate and control gear backlash in a reducer of a new energy vehicle electric drive system, including the following steps:
[0045] S1. Multi-dimensional factor data acquisition: Acquire multi-dimensional influencing factor data related to the gear clearance of the reducer in new energy vehicles. The multi-dimensional influencing factor data includes gear clearance reference data, motor torque data, reducer vibration reference data, temperature correlation coefficient data, mileage data, and gear force reference data.
[0046] S2. AI self-learning model processing: Input the multi-dimensional influencing factor data obtained in step S1 into the pre-trained AI self-learning model, and output the weight coefficients corresponding to each influencing factor through the AI self-learning model.
[0047] S3. Adaptive tooth-aligning force calculation: Based on the weight coefficients output in step S2 and the multi-dimensional influencing factor data obtained in step S1, an adaptive tooth-aligning force calculation formula is established, and the real-time optimal tooth-aligning force under the current working condition is obtained through the calculation formula.
[0048] S4. Dynamic compensation execution: Based on the real-time optimal gear force obtained in step S3, a command is sent to the motor controller to control the motor to output the corresponding torque, so that the meshing surface of the reducer gears remains in contact, thereby realizing real-time compensation of the reducer gear backlash.
[0049] S5. Model Iteration and Optimization: Continuously collect the transmission state data of the reducer after gear backlash compensation, and feed the transmission state data back to the AI self-learning model for iterative training, dynamically updating the weight coefficients corresponding to each influencing factor.
[0050] Specifically, this invention provides an intelligent self-learning and gear force compensation method for gear backlash in a new energy vehicle reducer. This method involves synchronously collecting multi-source sensor data on motor position, torque current, and reducer vibration from the electric drive system of the new energy electric vehicle; using a pre-trained artificial intelligence model to extract features from the data stream and perform condition adaptation analysis to dynamically identify the actual gear backlash value under the current driving conditions; and adjusting the backlash compensation parameters of the electric drive system controller in real time based on the identification results to achieve adaptive and precise compensation.
[0051] This invention breaks through the technical bottleneck of the traditional fixed-value compensation method. It can dynamically adapt to the time-varying and nonlinear characteristics of gear backlash as the driving mileage and road conditions change, significantly improving the transmission accuracy, driving smoothness and power response speed of the electric drive system. At the same time, it has the ability to monitor gear wear status and predictive maintenance, adapting to the driving needs of new energy electric vehicles under all working conditions, improving transmission smoothness and reliability, and ensuring that the NVH performance of the transmission system does not deteriorate.
[0052] In step S1 above, the gear backlash reference data is obtained in the following way:
[0053] During the reducer bench test and the vehicle EOL (Emergency Online Test) testing phases, the gear backlash self-learning command is initiated by the motor controller to control the motor output torque within ±5 N·m. The electrical angle change value of the drive motor during the switch from positive tooth surface engagement to negative tooth surface engagement of the reducer gear is recorded. The average value of the N collected electrical angle change values is taken to obtain the gear backlash reference value △θ. The specific steps are as follows:
[0054] S111. Calibration Preparatory Stage: During the bench test stage, the reducer under test and the motor are assembled to form the electric drive system assembly and fixed on the electric drive performance test bench. Communication debugging and signal calibration of the bench test and control system and the motor controller are completed. During the vehicle EOL off-line testing stage, the vehicle electric drive system assembly and vehicle power-on debugging are completed. The vehicle is in a parked and stationary state, the gear is switched to P, the handbrake is engaged, and the vehicle speed is 0 to ensure that there is no driving risk during the calibration process.
[0055] S112, Self-learning instruction trigger: The gear backlash self-learning instruction is sent through the motor controller MCU, and the MCU enters the backlash calibration control mode to control the output torque of the drive motor to perform bidirectional smooth loading within the range of ±5N.m;
[0056] S113. Electrical Angle Data Acquisition: First, control the motor to apply forward torque until the positive tooth surfaces of the reducer gear pair are completely rigidly engaged, locking the forward transmission path of the gear pair. Record the motor rotor electrical angle θ1 at this time. Then, control the drive motor to apply reverse torque until the negative tooth surfaces of the reducer gear pair are completely rigidly engaged, locking the reverse transmission path of the gear pair. Record the motor rotor electrical angle θ2 at this time. Calculate the electrical angle change value corresponding to the switching between the positive and negative tooth surfaces of the gear during a single calibration process = |θ2-θ1|.
[0057] S114. Reference value calculation: Repeat the acquisition process of step S113 at least 10 times, take the arithmetic mean of the effective single electrical angle change value, determine the average value as the gear backlash reference value △θ, and store it in the storage unit of the motor controller MCU as the initial reference parameter for subsequent tooth force compensation calculation.
[0058] Furthermore, to match the dynamic increase in gear clearance caused by long-term meshing wear of the reducer gear pair throughout the vehicle's entire life cycle, this invention correlates the gear clearance reference value Δθ with the vehicle's mileage data, and periodically and dynamically corrects Δθ to avoid the problem of compensation inaccuracy when the fixed reference value increases with mileage. In step S1 above, the gear clearance reference value Δθ is correlated and corrected with the mileage data, and the specific correction method is as follows:
[0059] Using 5000km as the step size, when the vehicle's cumulative mileage reaches the preset distance, under the condition of the vehicle's first power-down and meeting safe driving conditions, the gear backlash self-learning command is initiated through the motor controller. This controls the drive motor to output a torque of ±2N.m, records the electrical angle change value of the drive motor during the transition of the reducer gear from positive tooth surface engagement to negative tooth surface engagement, and corrects the gear backlash reference value △θ based on the electrical angle change value. The specific implementation process is as follows:
[0060] S121. Correction trigger condition: With 5000km as the fixed correction step, when the cumulative mileage M collected by the vehicle instrument system reaches an integer multiple of 5000km, the MCU automatically marks and triggers the gear backlash self-learning correction process.
[0061] S122 Correction mode calibration: After the triggering and safety conditions are met, the motor controller MCU automatically starts the gear backlash self-learning correction command, controls the drive motor to output a bidirectional torque of ±2 N.m, repeats the electrical angle acquisition process of step S113, obtains the single electrical angle change value corresponding to this correction process, repeats the acquisition for no less than 10 times, and takes the average value after processing in step S114 to obtain the actual backlash value corresponding to this correction.
[0062] S123, Reference value update: The measured value of the gear clearance obtained in this correction process is used as the new gear clearance reference value △θ, which overwrites the original reference value stored in the motor controller MCU, and the dynamic correction of the gear clearance reference value is completed. After the correction is completed, the new △θ is used as the reference parameter for subsequent calculations based on tooth force compensation until the next correction process is triggered.
[0063] The meshing clearance of the gear pair in the reducer will undergo dynamic elastic deformation with the change of load torque. The positive and negative directions of the torque directly correspond to the meshing state of the positive and negative tooth surfaces of the gear pair. The transient change of torque directly determines the timing of the gear reversal condition. Therefore, the acquisition accuracy, real-time performance and condition coverage of the torque command value T directly determine the adaptability and control accuracy of the tooth force compensation strategy of this invention.
[0064] In step S1 above, the motor torque data is the torque command value T of the motor controller. The torque command value T is read in real time through the vehicle controller (VCU) via the CAN bus, with a sampling frequency of not less than 100Hz. The torque command value T covers the torque signal of the entire vehicle driving, braking and coasting conditions. After the acquisition is completed, the torque signal is filtered and noise-reduced.
[0065] In this embodiment of the invention, the collected torque command value T fully covers the torque signal of the vehicle under all driving conditions. Under driving conditions, the torque command value T is positive, corresponding to the vehicle's forward and reverse acceleration, where the reducer gear pair is in positive tooth surface meshing. Under braking conditions, the torque command value T is negative, corresponding to the vehicle's regenerative braking, where the reducer gear pair is in negative tooth surface meshing. Under coasting conditions, the torque command value T approaches 0. Through full-coverage acquisition of torque signals under all driving conditions, it is ensured that the tooth force compensation strategy can adapt to the torque change requirements of all driving scenarios. The torque command value T directly reflects the trend of the vehicle's operating conditions and is a key feature parameter for the AI model to identify driving conditions and fit the mapping relationship between various factors and tooth force. Through full-condition, high-real-time torque data acquisition, accurate dynamic input is provided for the training and real-time computation of the AI model, ensuring the accuracy and adaptability of the AI model's output weight coefficients.
[0066] The reducer vibration reference value G is a characteristic parameter representing the gear meshing impact level and transmission NVH state of the reducer. It can directly reflect the severity of meshing impact and zero-crossing knocking noise caused by gear clearance, providing direct feature basis for optimizing the weight coefficients of the AI self-learning model. In step S1 above, the reducer vibration reference data is obtained in the following way:
[0067] During the end-of-life (EOL) testing phase of the vehicle, vibration data of the reducer within the 50Hz-8000Hz frequency band is collected by vibration sensors installed at predetermined key locations on the reducer housing. The root mean square value of the vibration data is calculated to obtain the reducer vibration reference value G. The vibration reference value G is stored in the motor controller MCU as a fixed input reference parameter for the AI self-learning model, and also as a comparison benchmark for subsequent compensation effect verification.
[0068] The core transmission components of the reducer, such as gears, bearings, and housing, are all made of metal. Their thermal deformation is directly related to the operating temperature. However, the reducer's operating oil temperature is not solely affected by ambient temperature but also by motor torque and speed. Existing fixed-value compensation methods completely fail to consider the dynamic changes in clearance under the coupled effects of temperature, load, and speed, leading to a severe deterioration in compensation performance under high and low temperature conditions. This invention introduces a temperature correlation coefficient Kt to quantify the impact of thermal deformation on clearance under multi-factor coupling, providing a temperature-dimensional correction basis for the AI self-learning model and achieving gear force compensation under all temperature conditions.
[0069] In step S1 above, the temperature correlation coefficient data is the linear relationship coefficient Kt between the reducer oil temperature and ambient temperature, motor torque, and motor speed. The linear relationship coefficient Kt is calculated in the following way:
[0070] Beforehand, the fitting constants d, e, and f are calibrated through a steady-state thermal test on the reducer bench, and a linear correlation model between oil temperature and ambient temperature, torque, and speed is established. During the real-time operation of the vehicle, real-time motor torque T, motor speed n, ambient temperature Tenv, and real-time reducer oil temperature Toil are collected, and the linear relationship coefficient Kt is calculated using the formula Kt=(Toil-Tenv) / (d*T+e*n+f).
[0071] The fitting constants d, e, and f are the fundamental parameters for constructing a linear correlation model between oil temperature and ambient temperature, torque, and speed. Their calibration accuracy directly determines the accuracy of the temperature correlation coefficient K_t, thus affecting the control accuracy relying on gear force compensation. In this invention, the standardization calibration of d, e, and f is completed through a steady-state thermal test on a reducer bench. The specific implementation steps are as follows:
[0072] S131. Pre-test Preparation: The reducer under test, the matching drive motor, and the dynamometer are coaxially assembled to form a complete electric drive system test assembly, which is then fixed to an electric drive performance test bench with closed-loop ambient temperature control capability. A temperature sensor is installed inside the reducer's oil sump to collect real-time oil temperature (Toil); an ambient temperature sensor is installed inside the test bench's environmental chamber to collect the test ambient temperature (Tenv); real-time communication is established between the test bench's measurement and control system and the motor controller MCU to synchronously collect motor torque (T) and motor speed (n) data. The sampling frequency of all data acquisition channels is not less than 10Hz to ensure the time synchronization of multi-source data. Before the test, lubricating oil of the same specification as that used in the mass-produced vehicle is added to the reducer, with the amount added consistent with that of the mass-produced vehicle, ensuring that the test boundary conditions are completely matched with the actual vehicle operating conditions.
[0073] S132. Full-condition test matrix setting: To cover all common operating scenarios throughout the entire vehicle lifecycle, a multi-dimensional gradient test condition matrix is constructed, with the gradient settings for each dimension as follows:
[0074] (1) Ambient temperature dimension: It covers the extreme scenarios of high and low temperature operation of the whole vehicle, and sets 7 gradient ambient temperature points of -40℃, -20℃, 0℃, 25℃, 40℃, 60℃ and 85℃, which fully covers the usage conditions of the whole scenario from extremely cold regions to high temperature environments;
[0075] (2) Motor torque dimension: Covering all working conditions from no load to heavy load, setting 6 gradient torque points: 0 N.m, 10% of rated torque, 25% of rated torque, 50% of rated torque, 75% of rated torque and 100% of rated torque, including full load operation scenarios such as coasting, constant speed cruising and heavy load climbing.
[0076] (3) Motor speed dimension: Covering all operating conditions from low speed to high speed of the whole vehicle, setting 5 gradient speed points of 1000rpm, 2000rpm, 4000rpm, 6000rpm and the maximum operating speed of the motor, fully covering the full speed range from low speed operating conditions in urban congestion to high speed cruising operating conditions.
[0077] The above working condition matrix contains a total of 210 test conditions to ensure that the calibration fitting constants can be adapted to all normal operating scenarios of the vehicle without any blind spots.
[0078] S133. Steady-state condition determination and effective data acquisition: For each set of test conditions, perform the test and data acquisition according to the following procedure:
[0079] (1) Preheating and temperature equalization: First, control the environmental chamber to stabilize to the preset ambient temperature point for this working condition, and keep it warm for no less than 2 hours to make the overall initial temperature of the electric drive system assembly completely consistent with the ambient temperature, thus eliminating the interference of the initial temperature deviation on the test results.
[0080] (2) Working condition loading: Control the drive motor to run continuously and stably with the preset torque T and speed n under the working condition, and monitor the change trend of reducer oil temperature Toil in real time during the operation;
[0081] (3) Thermal steady state determination: When the reducer oil temperature Toil meets the preset thermal steady state determination standard, that is, the oil temperature change rate is ≤0.5℃ / min within 5 consecutive minutes, the reducer is determined to have reached the thermal steady state under this working condition. At this time, the real-time oil temperature Toil, ambient temperature Tenv, motor torque T, and motor speed n of the reducer under this steady state are recorded simultaneously as a set of valid calibration data.
[0082] (4) Working condition reset: After the data acquisition of a single working condition is completed, the motor is stopped and the reducer oil temperature is cooled to a difference of ≤2℃ from the current ambient temperature before the next working condition test is performed to avoid the heat accumulation of the previous working condition causing systematic deviations in the subsequent test results;
[0083] S134. Data Fitting and Constant Calibration: After completing the tests for all 210 working conditions, all collected valid calibration data were processed and fitted. The specific process is as follows:
[0084] (1) Constructing a linear regression model: Using the difference between oil temperature and ambient temperature (Toil-Tenv) under each working condition as the dependent variable, and the motor torque T and motor speed n under the corresponding working condition as independent variables, a multiple linear regression model is constructed. The model expression is: Toil-Tenv = d•T + e•n + f;
[0085] (2) Linear fitting solution: The least squares method is used to perform linear fitting on all valid calibration data to obtain the fitting constants d, e, and f. During the fitting process, the 3σ criterion is used to remove abnormal data with residuals exceeding 3 times the standard deviation. After removal, the corresponding repeated test data of the working conditions are supplemented to ensure the effectiveness and completeness of the fitted dataset.
[0086] S135. Verification and Consolidation of Calibration Results: Substitute the fitted constants d, e, and f into the linear regression model above, and calculate and verify (Toil-Tenv) for all test conditions. The relative error between the model calculation value and the measured value under all conditions should be ≤5%. If there are conditions that exceed the error range, add gradient test points around the condition and refit until the verification error of all conditions meets the accuracy requirements. The finally determined fitting constants d, e, and f are stored in the motor controller MCU as fixed calibration parameters for Kt calculation during actual vehicle operation.
[0087] The meshing wear of the reducer gear pair is positively correlated with the vehicle's cumulative mileage. Existing fixed-value compensation methods completely fail to consider mileage-related wear changes, leading to a continuous deterioration of the compensation effect after long-distance vehicle operation, and even severe meshing impacts and abnormal noises. This invention introduces a mileage factor M and matches it with a dynamically adjusted weighting coefficient k5 to match the wear pattern throughout the gear's life cycle, achieving full-lifecycle adaptive optimization of the clearance compensation value. In step S1 above, the mileage data is the vehicle's cumulative mileage M, read in real-time from the vehicle's instrument system with an accuracy of no less than 1km; the weighting coefficient k5 corresponding to the vehicle's cumulative mileage M is dynamically adjusted according to the mileage interval.
[0088] When M≤5000km, the value of k5 ranges from 0.008 to 0.02;
[0089] When 5000km < M ≤ 80000km, the value of k5 ranges from 0.02 to 0.07.
[0090] When M > 80000 km, the value of k5 ranges from 0.07 to 0.11.
[0091] When the total mileage of the vehicle is M≤5000km, the gear pair is in the initial running-in stage, the wear rate of the tooth surface is slow, and the change in gear clearance is small. Therefore, the value range of k5 is set to 0.008-0.02, and a low weight coefficient is adopted to avoid the compensation value being too large due to excessive correction of the mileage item, so as to ensure the smoothness of transmission during the break-in period of the new car.
[0092] When 5000km < M ≤ 80000km, the gear pair completes the initial running-in and enters the stable wear stage, with a uniform and constant wear rate. Therefore, the value range of k5 is set to 0.02-0.07, using a medium weighting coefficient to match the stable growth trend of the gear clearance and ensure that the compensation value is adapted to the actual clearance.
[0093] When M > 80,000 km, the gear pair has been in service for a long time, and problems such as fatigue wear, microscopic pore corrosion, and spalling appear on the tooth surface. The wear rate is significantly accelerated, which is the high-incidence stage of gear meshing impact and knocking noise. Therefore, the value range of k5 is set to 0.07-0.11, and a high weighting coefficient is adopted to strengthen the correction effect of the mileage term on the tooth force compensation value, so as to ensure that the compensation value can accurately match the rapid change of clearance under long mileage and avoid compensation inaccuracy.
[0094] In step S1 above, the tooth force reference data is the tooth force reference value F, which is obtained through calibration via actual vehicle road testing. The specific calibration method is as follows:
[0095] The tooth force output of the drive motor is adjusted under various working conditions, including urban roads, highways, and mountain roads. The output range of the tooth force is ±5N, the control accuracy is no more than ±0.3N, and the response time is no more than 30ms. Combining subjective driving evaluation with objective test data of impact and vibration, the tooth force value corresponding to the optimal transmission smoothness of the reducer is determined as the tooth force reference value F.
[0096] The calibration based on the tooth force reference value F fully covers all driving conditions of the vehicle, including three major categories: urban road conditions, highway road conditions, and mountain road conditions. The specific calibration steps are as follows:
[0097] S141. Urban Road Condition Calibration: Calibration is carried out on standard urban roads, covering congested conditions, urban constant speed conditions, and reversing and turning conditions. Under the above conditions, the gear force output value is adjusted in a step size of 0.2N within a range of ±5N. Each gradient value is stably run for no less than 5 test cycles. Objective test data under the gradient is collected simultaneously, and the evaluation engineer completes the subjective scoring.
[0098] S142. Highway working condition calibration: Calibration is carried out on standard highways, using the same gradient adjustment method as urban working conditions, to complete objective data collection and subjective evaluation under each gear force gradient.
[0099] S143. Mountain Road Calibration: Calibration is carried out on standard mountain roads, covering uphill, downhill, and continuous curve conditions. The focus is on testing the transmission performance under scenarios of large load fluctuations and frequent torque reversal. The same gradient adjustment method is used to complete the full gradient test and data collection.
[0100] Next, the reference value F for tooth force is determined and solidified. The specific steps are as follows:
[0101] S151. Optimal value selection: After completing the calibration test for all working conditions, the test data under all gradients are processed. Among the effective gradient values, the tooth force value with the smallest objective test index and the highest subjective evaluation score is selected as the optimal tooth force value for the corresponding single working condition.
[0102] S152. Calculation of reference value: For the optimal tooth-feeding force value under all working conditions, after removing outliers using the 3σ criterion, the arithmetic mean is taken to obtain the final reference value F of tooth-feeding force.
[0103] S153. Verification and Consolidation: Write the calibrated tooth force reference value F into the motor controller MCU, and verify its reproduction under all calibration conditions to ensure that the transmission smoothness under all conditions meets the design requirements. After verification, solidify and store the tooth force reference value F in the motor controller MCU as a fixed reference parameter for adaptive tooth force calculation.
[0104] This invention uses an AI self-learning model as the core computing unit to fit the nonlinear mapping relationship between six input factors and the optimal gear backlash, and outputs the weight coefficients corresponding to each input factor. This addresses the shortcomings of existing technologies that rely on physical modeling to accurately characterize the time-varying and nonlinear distribution characteristics of gear backlash. In step S2 above, the AI self-learning model adopts a CNN-LSTM hybrid neural network architecture, which has higher fitting accuracy, faster convergence speed, and stronger generalization ability for various working conditions.
[0105] The input layer of the AI self-learning model standardizes the input multi-dimensional influencing factor data. The hidden layer includes two convolutional layers and three LSTM layers. The output layer outputs the weight coefficients k0-k5 corresponding to each influencing factor. The training convergence error of the AI self-learning model does not exceed 2%. The AI self-learning model is pre-trained with the optimal smoothness of the reducer transmission as the objective function, and learns the nonlinear mapping relationship between multi-dimensional influencing factor data and gear force.
[0106] In the AI self-learning model, the input layer has six input dimensions, corresponding to the six input factors of this invention: gear backlash reference value Δθ, motor torque command value T, reducer vibration reference value G, temperature correlation coefficient Kt, total vehicle mileage M, and gear force reference value F. The function of the input layer is to standardize and preprocess the multi-dimensional influencing factor data, eliminating the differences in dimensions and numerical ranges between different factors, and avoiding model fitting bias caused by inconsistent dimensions.
[0107] In the AI self-learning model, the output layer has 6 output dimensions, corresponding to 6 weight coefficients: the baseline compensation coefficient k0, the weight coefficient k1 corresponding to the gear backlash baseline value, the weight coefficient k2 corresponding to the motor torque command value, the weight coefficient k3 corresponding to the reducer vibration baseline value, the weight coefficient k4 corresponding to the temperature correlation coefficient, and the weight coefficient k5 corresponding to the vehicle's cumulative mileage. The output layer uses a linear activation function to ensure that the output weight coefficients can be directly substituted into the adaptive gear force calculation formula to complete the solution.
[0108] In step S3 above, based on the weight coefficients output in step S2 and the multi-dimensional influencing factor data obtained in step S1, an adaptive tooth-aligning force calculation formula is established. The real-time optimal tooth-aligning force under the current working condition is obtained through the calculation formula. The adaptive tooth-aligning force calculation formula is as follows:
[0109]
[0110] Wherein, Fk is the real-time optimal gear force, F is the gear force reference value, △θ is the gear backlash reference value, T is the torque command value of the motor controller MCU, G is the reducer vibration reference value, Kt is the temperature correlation coefficient, M is the vehicle's cumulative mileage, k0 is the reference compensation coefficient, and k1-k5 are the weight coefficients corresponding to each influencing factor output by the AI self-learning model.
[0111] In each control cycle, the motor controller MCU first completes the real-time acquisition and update of six core input factors. Then, it outputs the weight coefficients k0-k5 corresponding to the current operating condition through a fixed AI self-learning model. Subsequently, these coefficients are substituted into the adaptive gear force calculation formula to obtain the real-time optimal gear force Fk for the current control cycle. In step S4 above, based on the real-time optimal gear force obtained in step S3, a command is issued to the motor controller to control the motor to output the corresponding torque, keeping the meshing surfaces of the reducer gears in contact, completely eliminating the transmission backlash caused by gear clearance, and realizing real-time compensation for reducer gear clearance.
[0112] During the transient process of motor torque commutation and gear meshing surface switching, the motor controller predicts the commutation action in advance by changing the trend of torque command value, dynamically adjusts the gear force output, and ensures that there is no backlash or idle travel during the entire gear surface switching process, thereby suppressing gear zero-crossing knocking noise and meshing impact from the root.
[0113] In step S5 above, the reducer transmission status data includes the compensated reducer vibration data, drive motor torque data, drive motor position data, and reducer oil temperature data. The AI self-learning model continuously iterates based on the transmission status data, so that the compensation accuracy continuously adapts to the changes in the vehicle's operating conditions and the wear status of the reducer gears.
[0114] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution; or the direct application of the inventive concept and technical solution to other situations without modification, are all within the protection scope of the present invention.
Claims
1. A method for compensating for gear backlash and tooth force in a reducer of a new energy vehicle, characterized in that, Including the following steps: S1. Multi-dimensional factor data acquisition: Acquire multi-dimensional influencing factor data related to the gear clearance of the reducer in new energy vehicles. The multi-dimensional influencing factor data includes gear clearance reference data, motor torque data, reducer vibration reference data, temperature correlation coefficient data, mileage data, and gear force reference data. S2. AI self-learning model processing: Input the multi-dimensional influencing factor data obtained in step S1 into the pre-trained AI self-learning model, and output the weight coefficients corresponding to each influencing factor through the AI self-learning model. S3. Adaptive tooth-aligning force calculation: Based on the weight coefficients output in step S2 and the multi-dimensional influencing factor data obtained in step S1, an adaptive tooth-aligning force calculation formula is established, and the real-time optimal tooth-aligning force under the current working condition is obtained through the calculation formula. S4. Dynamic compensation execution: Based on the real-time optimal gear force obtained in step S3, a command is sent to the motor controller to control the motor to output the corresponding torque, so that the meshing surface of the reducer gears remains in contact, thereby realizing real-time compensation of the reducer gear backlash. S5. Model Iteration and Optimization: Continuously collect the transmission state data of the reducer after gear backlash compensation, and feed the transmission state data back to the AI self-learning model for iterative training, dynamically updating the weight coefficients corresponding to each influencing factor.
2. The method according to claim 1, characterized in that, In step S1, the gear backlash reference data is obtained in the following manner: During the reducer bench test and the vehicle EOL off-line testing, the gear backlash self-learning command is initiated by the motor controller to control the motor output torque within ±5 N.m. The electrical angle change value of the drive motor during the process of the reducer gear switching from positive tooth surface contact to negative tooth surface contact is recorded. The average value of the electrical angle change value collected N times is taken to obtain the gear backlash reference value △θ.
3. The method according to claim 2, characterized in that, The gear clearance reference value △θ is correlated and corrected with the mileage data, and the specific correction method is as follows: Using 5000km as the step size, when the cumulative mileage of the vehicle reaches the preset number of kilometers, under the condition that the vehicle is powered down for the first time and the safe driving conditions are met, the gear backlash self-learning command is initiated through the motor controller to control the drive motor to output a torque of ±2N.m, record the change value of the electric angle of the drive motor during the process of the reducer gear switching from positive tooth surface contact to negative tooth surface contact, and correct the gear backlash reference value △θ based on the change value of the electric angle.
4. The method according to any one of claims 1 to 3, characterized in that, In step S1, the motor torque data is the torque command value T of the motor controller. The torque command value T is read in real time through the vehicle controller via the CAN bus, with a sampling frequency of not less than 100Hz. The torque command value T covers the torque signal of the entire vehicle under all driving, braking, and coasting conditions. After the acquisition is completed, the torque signal is filtered and denoised.
5. The method according to any one of claims 1 to 3, characterized in that, In step S1, the vibration reference data of the reducer is obtained in the following manner: During the end-of-life (EOL) testing phase of the vehicle, vibration data of the reducer in the 50Hz-8000Hz frequency band is collected by vibration sensors installed at preset key positions on the reducer housing. The root mean square value of the vibration data is calculated to obtain the reducer vibration reference value G.
6. The method according to any one of claims 1 to 3, characterized in that, In step S1, the temperature correlation coefficient data is the linear relationship coefficient Kt between the reducer oil temperature and ambient temperature, motor torque, and motor speed. The linear relationship coefficient Kt is calculated in the following way: Beforehand, the fitting constants d, e, and f are calibrated through a steady-state thermal test on the reducer bench, and a linear correlation model between oil temperature and ambient temperature, torque, and speed is established. During the real-time operation of the vehicle, real-time motor torque T, motor speed n, ambient temperature Tenv, and real-time reducer oil temperature Toil are collected, and the linear relationship coefficient Kt is calculated using the formula Kt=(Toil-Tenv) / (d*T+e*n+f).
7. The method according to any one of claims 1 to 3, characterized in that, In step S1, the mileage data is the vehicle's cumulative mileage M, which is read in real time from the vehicle's instrument system with an accuracy of not less than 1km; the weighting coefficient k5 corresponding to the vehicle's cumulative mileage M is dynamically adjusted according to the mileage interval. When M≤5000km, the value of k5 ranges from 0.008 to 0.02; When 5000km < M ≤ 80000km, the value of k5 ranges from 0.02 to 0.
07. When M > 80000 km, the value of k5 ranges from 0.07 to 0.
11.
8. The method according to any one of claims 1 to 3, characterized in that, In step S1, the tooth-mounting force reference data is the tooth-mounting force reference value F, which is obtained through calibration via actual vehicle road testing. The specific calibration method is as follows: The tooth force output of the drive motor is adjusted under various working conditions, including urban roads, highways, and mountain roads. The output range of the tooth force is ±5N, the control accuracy is no more than ±0.3N, and the response time is no more than 30ms. Combining subjective driving evaluation with objective test data of impact and vibration, the tooth force value corresponding to the optimal transmission smoothness of the reducer is determined as the tooth force reference value F.
9. The method according to any one of claims 1 to 3, characterized in that, The AI self-learning model adopts a CNN-LSTM hybrid neural network architecture. The input layer of the AI self-learning model standardizes the input multi-dimensional influencing factor data. The hidden layer includes two convolutional layers and three LSTM layers. The output layer outputs the weight coefficients k0-k5 corresponding to each influencing factor. The training convergence error of the AI self-learning model does not exceed 2%. The AI self-learning model is pre-trained with the optimal smoothness of the reducer transmission as the objective function, and learns the nonlinear mapping relationship between multi-dimensional influencing factor data and gear force.
10. The method according to any one of claims 1 to 3, characterized in that, In step S3, the adaptive tooth-aligning force calculation formula is: Wherein, Fk is the real-time optimal gear force, F is the gear force reference value, △θ is the gear backlash reference value, T is the torque command value of the motor controller MCU, G is the reducer vibration reference value, Kt is the temperature correlation coefficient, M is the vehicle's cumulative mileage, k0 is the reference compensation coefficient, and k1-k5 are the weight coefficients corresponding to each influencing factor output by the AI self-learning model.
