An energy efficiency optimization method for a dual-motor servo press based on edge computing

Abstract

This invention relates to the field of motor servo control technology and discloses an energy efficiency optimization method for a dual-motor servo press based on edge computing. The method involves deploying an edge computing node on the dual-motor servo press body to collect real-time operating parameters of the main drive motor and auxiliary motor; dynamically constructing a dual-motor collaborative load state matrix based on the operating parameters; calculating a real-time comprehensive energy efficiency index using an energy efficiency evaluation model in the edge computing node; determining the dynamic energy efficiency adjustment parameters of the dual motors based on the deviation of this index from the target energy efficiency threshold; sending instructions from the edge computing node to the controller to synchronously update the power allocation strategy; and cyclically executing monitoring and feedback adjustments until the dual-motor collaborative load state reaches a preset energy efficiency stability range. This method achieves real-time monitoring and dynamic optimization of the dual-motor collaborative operation state, improving the energy efficiency stability of the equipment operation.

Description

Technical Field

[0001] This invention relates to the field of motor servo control technology, specifically to an energy efficiency optimization method for a dual-motor servo press based on edge computing. Background Technology

[0002] Dual-motor servo presses, as key equipment in modern industrial production, are widely used in automobile manufacturing, aerospace, precision instruments, and other fields. Their operational energy efficiency directly affects production cost control and environmental performance. With the continuous improvement of industrial automation, the load conditions of dual-motor servo presses are becoming increasingly complex, and the coordinated operation of the main drive motor and auxiliary motor has a more and more significant impact on overall energy efficiency.

[0003] Currently, traditional energy efficiency optimization methods mostly rely on preset control parameters or offline data analysis, making it difficult to respond to dynamic changes in load in real time. For example, during the stamping process, fluctuations in material thickness and hardness can cause instantaneous changes in motor load. Traditional control systems often adjust based on fixed power distribution strategies, which cannot adapt to dynamic load fluctuations in a timely manner, resulting in energy efficiency losses.

[0004] When two motors operate in tandem, the rationality of load distribution between the main drive motor and the auxiliary motor directly affects overall energy efficiency. An imbalanced load distribution may cause one motor to operate in an inefficient range for an extended period, or even experience overload, reducing energy efficiency and impacting equipment lifespan. Furthermore, traditional energy efficiency assessments are often based on the operating parameters of a single motor, lacking a comprehensive consideration of the tandem state of the two motors. This makes it difficult to accurately reflect the overall energy efficiency level of the equipment, thus affecting the effectiveness of optimization strategies.

[0005] With the development of Industrial Internet of Things (IIoT) technology, real-time data acquisition and analysis have become crucial for improving equipment energy efficiency. However, traditional cloud computing models suffer from data transmission latency and high bandwidth consumption when processing data with high real-time requirements, making it difficult to meet the real-time dynamic adjustment needs of dual-motor servo presses. Therefore, how to achieve collaborative energy efficiency assessment and dynamic optimization of dual motors based on real-time data has become a problem that needs to be solved. Summary of the Invention

[0006] The purpose of this invention is to provide an energy efficiency optimization method for a dual-motor servo press based on edge computing, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, this invention provides an energy efficiency optimization method for a dual-motor servo press based on edge computing, the method comprising:

[0008] The operating parameters of the main drive motor and the auxiliary motor are collected in real time by the edge computing node deployed on the dual-motor servo press body;

[0009] The dual-motor collaborative load state matrix is ​​dynamically constructed based on operating parameters, and the real-time comprehensive energy efficiency index is calculated through the energy efficiency evaluation model in the edge computing node.

[0010] Based on the deviation between the real-time comprehensive energy efficiency index and the target energy efficiency threshold, the dynamic energy efficiency adjustment parameters of the main drive motor and the auxiliary motor are determined.

[0011] The edge computing node sends dynamic energy efficiency adjustment parameter commands to the controller of the dual-motor servo press, and synchronously updates the power allocation strategy of the main drive motor and the auxiliary motor.

[0012] The system continuously monitors the real-time comprehensive energy efficiency index and adjusts the dynamic energy efficiency regulation parameters until the dual-motor coordinated load state reaches the preset energy efficiency stability range.

[0013] Preferably, the operating parameters include the real-time output torque of the main drive motor, the instantaneous speed of the auxiliary motor, the current phase difference when the two motors work together, and the displacement acceleration of the press slide.

[0014] Preferably, the dynamic construction of the dual-motor cooperative load state matrix includes:

[0015] Extract the torque fluctuation coefficient of the main drive motor and the speed fluctuation coefficient of the auxiliary motor within a preset time window;

[0016] Calculate the coupling influence factor between the phase difference of the dual motor currents and the acceleration of the slider displacement;

[0017] The torque fluctuation coefficient, speed fluctuation coefficient, and coupling influence factors are integrated into a multi-dimensional state matrix according to the time series.

[0018] Preferably, the calculation process of the real-time comprehensive energy efficiency index includes:

[0019] A multidimensional state matrix is ​​input into the energy efficiency assessment model, and the load characteristics of the main drive motor and the auxiliary motor are separated by a spatiotemporal feature extraction network.

[0020] By integrating load characteristics with preset motor power loss baseline values, dynamic energy efficiency parameters are generated for dual-motor collaborative operation.

[0021] The dynamic energy efficiency parameters are normalized and weighted to output a real-time comprehensive energy efficiency index in the range of 0-1.

[0022] Preferably, determining the dynamic energy efficiency adjustment parameters of the main drive motor and the auxiliary motor includes:

[0023] Establish an energy efficiency change curve with real-time comprehensive energy efficiency index as the vertical axis and timestamp as the horizontal axis;

[0024] Identify the periods in the energy efficiency change curve where the energy efficiency is continuously below the target energy efficiency threshold, and extract the maximum torque fluctuation coefficient and minimum speed fluctuation coefficient corresponding to those periods.

[0025] Based on the ratio between the maximum torque fluctuation coefficient and the standard torque threshold, the torque compensation amount of the main drive motor is generated.

[0026] The speed compensation amount for the auxiliary motor is generated based on the difference between the minimum speed fluctuation coefficient and the reference speed.

[0027] Preferably, the update of the power allocation strategy includes:

[0028] The torque compensation is added to the current output torque command value of the main drive motor;

[0029] The speed compensation is embedded in the closed-loop control feedback loop of the auxiliary motor;

[0030] The updated dual-motor current phase difference is verified using edge computing nodes to ensure it meets the preset collaborative tolerance range.

[0031] Preferably, the feedback adjustment process includes:

[0032] After the power allocation strategy is updated, the slider displacement acceleration during the update period is re-acquired;

[0033] Calculate the rate of change of variance of the slider displacement acceleration before and after the update;

[0034] If the variance change rate exceeds the preset threshold, the adjustment range of torque compensation and speed compensation will be reduced proportionally.

[0035] Preferably, the determination of the energy efficiency stability range includes:

[0036] Continuously monitor the real-time comprehensive energy efficiency index for three consecutive time windows;

[0037] When the standard deviation of the real-time comprehensive energy efficiency index in the three time windows is less than the preset tolerance value and is higher than 90% of the target energy efficiency threshold, it is determined that the energy efficiency has entered the stable range.

[0038] Preferably, when performing energy efficiency optimization, the edge computing node simultaneously compresses the dynamic energy efficiency adjustment parameters and corresponding operating parameters into encrypted data packets, and transmits them to the cloud server via an industrial IoT protocol for long-term performance analysis.

[0039] Preferably, the continuous monitoring process for the energy efficiency stability range includes:

[0040] When the standard deviation of the real-time comprehensive energy efficiency index exceeds the preset tolerance value, the preceding data of the current dual-motor collaborative load state matrix is ​​retrieved from the historical database of the edge computing node.

[0041] Extract historical records from the preceding data that match the current torque fluctuation coefficient and speed fluctuation coefficient, and calculate the mean variance of the slider displacement acceleration corresponding to the historical records;

[0042] The current variance of slider displacement acceleration is compared with the historical mean variance. If the deviation exceeds the threshold, the initialization and reset of the dynamic energy efficiency adjustment parameters are triggered.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] By deploying edge computing nodes on the dual-motor servo press body, real-time acquisition of operating parameters of the main drive motor and auxiliary motor is achieved. This overcomes the data transmission latency limitations of traditional cloud computing models and enables rapid response to changes in equipment operating status. Based on the real-time acquired operating parameters, a dual-motor collaborative load status matrix is ​​dynamically constructed, incorporating the operating status of both motors into a unified analysis framework. Compared to traditional evaluation methods that only consider single motor parameters, this provides a more comprehensive reflection of the load distribution when the two motors work collaboratively, offering a more accurate basis for subsequent energy efficiency assessments.

[0045] The energy efficiency assessment model integrated into the edge computing node can calculate a real-time comprehensive energy efficiency index based on the collaborative load state matrix. This index comprehensively considers the operating characteristics and collaborative relationship of the two motors, making the energy efficiency assessment results more accurate and comprehensive. Dynamic energy efficiency adjustment parameters are determined based on the deviation between the real-time comprehensive energy efficiency index and the target energy efficiency threshold, enabling the adjustment strategy to accurately match the current energy efficiency status of the equipment and avoiding the blindness of traditional fixed-parameter adjustment methods.

[0046] By sending adjustment parameter commands directly to the controller from edge computing nodes, rapid transmission and execution of these commands are achieved, ensuring timely dynamic energy efficiency adjustments. This allows for immediate intervention when equipment operating conditions fluctuate, maintaining the high efficiency of the dual-motor operation. The cyclical monitoring and feedback adjustment process forms a continuously optimized closed loop, ensuring that the dual-motor coordinated load state remains stable within the preset energy efficiency range. This avoids energy efficiency fluctuations caused by external factors or changes in the equipment's own condition, guaranteeing the long-term energy efficiency stability of the equipment.

[0047] The dynamically updated power allocation strategy can flexibly adjust the power output of both motors based on their real-time load conditions, ensuring that the main drive motor and auxiliary motor always operate within their respective high-efficiency ranges. This reduces energy loss caused by improper load allocation and lowers the likelihood of a single motor operating in an inefficient or overloaded state for extended periods, thus helping to extend the equipment's lifespan. This dynamic optimization mode based on real-time data can adapt to different load conditions, maintaining good energy efficiency under both stable and dynamic load environments, thereby improving the adaptability of the dual-motor servo press in complex industrial scenarios. Attached Figure Description

[0048] Figure 1 This is a timing diagram of the energy efficiency optimization method for a dual-motor servo press based on edge computing described in this invention.

[0049] Figure 2 Flowchart for constructing the dual-motor cooperative load state matrix;

[0050] Figure 3 Flowchart for determining dynamic energy efficiency adjustment parameters;

[0051] Figure 4 The flowchart was adjusted based on feedback.

[0052] Figure 5 A flowchart for continuous monitoring of energy efficiency within a stable range. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Please see Figure 1 This invention provides an energy efficiency optimization method for a dual-motor servo press based on edge computing, the method comprising:

[0055] The operating parameters of the main drive motor and auxiliary motor are collected in real time by edge computing nodes deployed on the dual-motor servo press body. A dual-motor collaborative load state matrix is ​​dynamically constructed, and a real-time comprehensive energy efficiency index is calculated based on an energy efficiency assessment model. Based on the deviation between the real-time comprehensive energy efficiency index and the target energy efficiency threshold, dynamic energy efficiency adjustment parameters for the main drive motor and auxiliary motor are determined. These dynamic energy efficiency adjustment parameter commands are sent to the controller of the dual-motor servo press via the edge computing nodes, synchronously updating the power allocation strategy for the main drive motor and auxiliary motor. The monitoring of the real-time comprehensive energy efficiency index and the feedback adjustment of the dynamic energy efficiency parameters are executed cyclically until the dual-motor collaborative load state reaches the preset energy efficiency stability range.

[0056] Example 1: See Figure 2 The energy efficiency optimization of the dual-motor servo press relies on the precise acquisition and analysis of operating parameters. The real-time output torque of the main drive motor is acquired through a high-precision torque sensor. This sensor is integrated at the mechanical connection between the motor output shaft and the driven component, directly measuring the torsional stress during transmission. The torque signal sampling frequency is no less than 1,000 times per second to ensure the capture of rapid dynamic changes. The instantaneous speed of the auxiliary motor is measured in real-time by a photoelectric encoder mounted on the motor rotor shaft. The encoder uses a high-resolution model, and the output signal is a differential pulse sequence. The edge computing node captures the pulse edges through a high-speed counter and calculates the instantaneous angular velocity using a precise clock reference, with measurement accuracy controlled within ±0.1 revolution per minute. The current phase difference during dual-motor collaborative operation requires the synchronous acquisition of the three-phase stator current signals of both motors. After the current signal is transformed into a suitable sampling voltage signal by an isolated current transformer, it is input to a multi-channel synchronous sampling analog-to-digital converter. The edge computing node performs fast Fourier transform analysis on the acquired current waveform data to accurately calculate the phase angle of the fundamental current, thereby obtaining the real-time phase difference angle between the current signals of the two motors. The displacement acceleration of the press slide is acquired by a triaxial piezoelectric accelerometer rigidly connected to the slide body. The output signal of this sensor is conditioned by a charge amplifier and then input to the analog acquisition module. The sampling frequency setting is strictly synchronized with the motor parameter acquisition to ensure the consistency of spatiotemporal data.

[0057] The dynamic construction of the dual-motor coordinated load state matrix is ​​a process of multi-dimensional feature extraction and integration. The system sets a preset time window length, which must cover an integer multiple of the typical working cycle of the press, such as 200 milliseconds. Within this time window, the torque fluctuation coefficient of the main drive motor is determined through calculation. Specifically, this involves statistically analyzing all torque sampling points within the window, calculating their standard deviation, and then dividing this standard deviation by the average torque value within the window; the resulting ratio is defined as the torque fluctuation coefficient. This coefficient objectively reflects the stability of the torque output. The speed fluctuation coefficient of the auxiliary motor uses a similar calculation logic. First, the standard deviation of all speed sampling values ​​within the time window is calculated; then, this standard deviation is divided by the average speed value of the corresponding time window; the resulting ratio is the speed fluctuation coefficient, characterizing the stability level of speed control. There is an inherent energy coupling relationship between the dual-motor current phase difference and the slider displacement acceleration; the calculation of its coupling influence factor requires comprehensive consideration of both sets of parameters. A mathematical model is established using multiple regression analysis to determine the relationship between the change in current phase difference and the change in slider displacement acceleration. The regression coefficients of this model serve as the coupling influence factor, which quantitatively describes the combined impact of these two changes on the overall energy efficiency of the system. The three characteristic quantities mentioned above—torque fluctuation coefficient, speed fluctuation coefficient, and coupling influence factor—must be integrated according to a strict time series. The integration method constructs a multidimensional state matrix with fixed dimensions. Each row of the matrix corresponds to a sampling time point, and the number of rows equals the total number of sampling points within the preset time window. The matrix contains three columns of data, storing the calculated values ​​of the torque fluctuation coefficient, speed fluctuation coefficient, and coupling influence factor at that specific time point. This matrix is ​​implemented in memory using a circular buffer and is updated in real time.

[0058] The accuracy of operational parameter acquisition directly affects the reliability of the load state matrix. Torque sensor calibration requires static calibration and dynamic frequency response testing using a standard weight lever device to ensure that amplitude error and phase shift within the motor's operating frequency band meet technical specifications. The installation of the photoelectric encoder must strictly ensure coaxiality with the motor shaft and eliminate gap errors in the connection device. The encoder signal interface circuit design employs anti-interference measures, such as twisted-pair shielded cable transmission and Schmitt trigger shaping at the receiving end, to reduce the impact of environmental electromagnetic noise. The selection of the current transformer's core material and turns ratio design must balance bandwidth and accuracy, paying particular attention to nonlinear distortion characteristics under large dynamic loads. Current signal sampling uses a multi-channel synchronous hold analog-to-digital converter with a conversion resolution of at least 16 bits and a conversion speed sufficient to support accurate restoration of the fundamental frequency. The piezoelectric accelerometer must be selected based on the measurement frequency band and range, and high-strength bolts must be used for fixing to ensure that the sensor's fundamental frequency is much higher than the upper limit of the measured signal. The charge amplifier needs to be configured with a high-pass filter with an appropriate time constant to eliminate zero-point drift and set with an appropriate sensitivity range to adapt to changes in acceleration amplitude under different operating conditions. All these sensor signals undergo front-end conditioning via necessary anti-aliasing low-pass filters before entering the central processing unit of the edge computing node.

[0059] Edge computing nodes perform preliminary preprocessing on the acquired raw sampled data. Raw torque and speed data may contain sudden interference spikes, so a sliding median filter algorithm is used for smoothing. The filter window width is set to a number of sampling points, which must be greater than the typical interference pulse width. Spectral characteristic analysis of the slider displacement acceleration signal is crucial; therefore, a Fast Fourier Transform (FFT) is performed within the edge node to analyze the distribution of the main frequency components in the signal. This information helps in subsequent determination of the equipment's mechanical resonance point and load characteristics. Timestamps of operating parameters are used in conjunction with the edge node's built-in real-time clock chip and the operating system's precise time protocol service to ensure microsecond-level time alignment accuracy for multi-source data streams. Preprocessed key parameters are categorized and stored in the edge node's database partition. This database is organized using a time-series data table structure, with each parameter allocated independent data channel storage space based on its physical characteristics and update frequency. The database provides a high-efficiency query interface, supporting the extraction of historical data records by time range or specific operating condition tags. After the data storage period ends, older records exceeding a certain time limit can be automatically archived to non-volatile storage media or deleted according to a policy to manage storage space resources.

[0060] The construction of the multidimensional state matrix is ​​dynamic and continuous. The central processing unit of the edge computing node continuously schedules a dedicated matrix generation task. This task first loads the preprocessed raw data of all relevant parameters within a set time window from the database. The loading operation must strictly ensure the temporal integrity of the data. Next, the task starts the torque fluctuation coefficient calculation module. This module loads the raw sample array of the main drive motor torque within the time window, calculates the arithmetic mean of the array elements as a reference, and then calculates the sum of squares of the deviations of the torque value of each sample point from this average. The sum of squares of the deviations is divided by the number of sample points minus one to obtain the variance value. The square root of the variance value is the standard deviation of the torque. Finally, the standard deviation of the torque is divided by the average torque value to output the torque fluctuation coefficient. The speed fluctuation coefficient calculation uses the exact same algorithm logic to independently process the auxiliary motor speed data and outputs the speed fluctuation coefficient value. The calculation process of the coupling influence factor is relatively complex: this module loads synchronously acquired current phase difference time series data and slider displacement acceleration time series data. Covariance analysis is performed on the two sets of data to calculate their covariance values. The variance values ​​of the current phase difference data and the slider displacement acceleration data are calculated separately. Finally, the covariance value is divided by the product of the standard deviations of the current phase difference and the slider displacement acceleration to obtain the Pearson correlation coefficient, which characterizes the linear correlation between the two sets of parameters. This coefficient is a core component of the coupling influence factor. Within the same processing task cycle, the system, based on the current time index point, sequentially fills the current row of the multidimensional state matrix with the real-time calculated torque fluctuation coefficient, speed fluctuation coefficient, and coupling influence factor values. The matrix row index increments forward over time. When the processing task cycle ends, a complete multidimensional state matrix containing recent load dynamic characteristics is constructed and can be used by downstream energy efficiency assessment models. The matrix construction cycle is a configurable parameter of the system, typically matching the typical operating rhythm of the press. The matrix data is temporarily stored in a high-speed shared memory area and simultaneously written to a historical state database on a non-volatile storage device for persistent storage.

[0061] State matrix management includes necessary lifecycle control. Newly constructed matrices overwrite the oldest historical matrix copies, enabling cyclical use of storage space. The system generates a unique index identifier and detailed metadata for each round of matrix construction, including construction timestamps, the covered time window range, the total number of collected data points, and a feature value statistical overview. In the non-volatile storage portion, the state matrix is ​​structured and stored as binary large object data blocks combined with metadata records. The external data service interface of the edge computing nodes supports querying and downloading load state matrix data for specific time periods, facilitating offline analysis or remote diagnostics. The matrix data compression algorithm uses highly efficient lossless compression methods, such as dictionary-based compression algorithms, saving storage space and network transmission bandwidth while maintaining data accuracy. When edge nodes need to store massive amounts of matrix data long-term, a time-sharing archiving strategy is used to migrate earlier data to larger-capacity storage media. This dynamic construction, efficient storage, and cyclical utilization mechanism ensures the data resource management required for the long-term stable operation of the system.

[0062] Example 2: See Figure 3 After the multidimensional state matrix is ​​generated, it enters the calculation process of the real-time comprehensive energy efficiency index. The core of the energy efficiency assessment model is the spatiotemporal feature extraction network. This network adopts a parallel processing architecture, with the input layer receiving multidimensional state matrix data organized in time series. The first part of the network is a one-dimensional convolutional layer, which is configured with multiple convolutional kernels of different scales sliding along the time axis. These convolutional kernels capture the local change patterns of torque fluctuation coefficients, speed fluctuation coefficients, and coupling influencing factors in the state matrix, such as short-term load mutations or periodic oscillation signals. The convolutional output is processed by a nonlinear activation function and then fed into a pooling layer. The pooling layer selects the most significant feature responses, retains their positional information, and reduces the data dimensionality. The second part of the network consists of a long short-term memory structure. The long short-term memory unit receives the feature sequence from the convolutional layer and uses its internal gating mechanism to learn the long-term temporal dependencies of state changes. The forget gate determines the degree of retention of previous memory states, the input gate adjusts the fusion weights of new features, and the output gate controls the release of information from the memory unit to subsequent network layers. The output of the long short-term memory network carries the temporal evolution characteristics of the load state.

[0063] This network employs a branching structure to separate the load characteristics of the main drive motor and the auxiliary motor. After the Long Short-Term Memory (LSTM) layer, the network branches into two independent sub-paths. Each path contains a fully connected layer for deep feature reorganization. Through weight configuration, one path focuses on the load correlation pattern of the main drive motor, while the other focuses on the load characteristics of the auxiliary motor. Each path outputs a high-dimensional vector containing specific load characteristics, which abstractly represents the energy conversion characteristics of the corresponding motor under the current operating state. The fully connected layers use different weight parameter initialization schemes to encourage the two paths to learn differentiated feature representations.

[0064] These load feature vectors are fused with a preset baseline value for motor power loss. The baseline value for power loss, derived from motor design specifications and offline test data, comprises three components: copper loss, iron loss, and mechanical loss. Copper loss is calculated based on the DC resistance and rated current of the motor windings; iron loss includes eddy current loss and hysteresis loss, determined based on the stator core material properties and operating magnetic flux density; mechanical loss includes calculated values ​​of bearing friction loss and wind resistance loss. The fusion operation is achieved through feature splicing, which connects the load feature vector with the three loss component values ​​to form an extended feature vector. This fusion process allows the model to correlate real-time load dynamics with inherent energy consumption characteristics.

[0065] The extended feature vector is input to another set of fully connected layers for feature mapping. This layer learns the nonlinear relationship between load dynamics and loss components, outputting a transient power loss parameter. This parameter represents the real-time power conversion efficiency of the system under a specific load condition. The model further maps this transient parameter to the interval of 0 to 1 through a normalization transformation. The normalization reference coordinate system is dynamically constructed based on historical operating data, with the minimum value taken as the minimum transient power loss value detected within the most recent time window, and the maximum value taken as the maximum loss value within that time period. The transient power loss parameter is compressed into this interval through a linear transformation. Since the energy consumption contribution of the main drive motor and the auxiliary motor differs in different working stages of the press, the system sets dynamic weighting coefficients for the final synthesis of the comprehensive energy efficiency index. The weighting coefficients are automatically adjusted according to the stroke position signal of the press controller; for example, the weight of the auxiliary motor is increased during the acceleration stage, and the weight of the main drive motor is increased during the stamping stage. Finally, the real-time comprehensive energy efficiency index is calculated by weighted average, with the value closer to 1 indicating a higher overall system energy efficiency level.

[0066] The process of determining dynamic energy efficiency adjustment parameters begins with the analysis of energy efficiency change curves. The system continuously records timestamps and corresponding real-time comprehensive energy efficiency index values, maintaining a fixed-length energy efficiency data sequence in memory. This sequence forms a time trajectory graph of energy efficiency changes. The system applies a sliding window detection algorithm to scan this curve, with the window length set to the duration of a typical abnormal state. The detection logic identifies whether all energy efficiency index data points in the window are below a preset target energy efficiency threshold. When three or more consecutive sampling points are detected to be below this threshold, the system determines that there is an energy efficiency anomaly period and locks that period for in-depth analysis.

[0067] During this abnormal period, the system performs feature extraction. All main drive motor torque fluctuation coefficient data points are retrieved from the multi-dimensional state matrix slice corresponding to this period. The maximum value is recorded as the maximum torque fluctuation coefficient by scanning these points. Similarly, the auxiliary motor speed fluctuation coefficient data sequence is retrieved during this period, and the minimum speed fluctuation coefficient is determined after eliminating random fluctuation interference using a moving average filter. These extreme values ​​reflect the most significant load imbalance state during the abnormal period.

[0068] The determination of the torque compensation amount relies on a pre-calibrated relational model. The system maintains a pre-calibrated torque-energy efficiency response surface data table, which describes the sensitivity of energy efficiency indicators to changes under different torque ripple coefficients. The ratio of the current maximum torque ripple coefficient to the standard torque threshold is calculated. The standard torque threshold is preset based on the press's rated load and motor characteristics. The energy efficiency loss ratio corresponding to this ratio is obtained by querying the response surface, and the required torque compensation amount is derived accordingly. A positive compensation amount indicates that the torque output ratio of the main drive motor needs to be increased, while a negative compensation amount indicates that the torque output ratio needs to be reduced.

[0069] The calculation of speed compensation employs a difference processing logic. First, the reference speed value of the auxiliary motor is retrieved. This value comes from the press process parameter database and is bound to the current processing code. The absolute difference between the detected minimum speed fluctuation coefficient and the reference speed is calculated. Since the speed fluctuation coefficient is a relative ratio rather than a physical speed value, it needs to be converted to the corresponding physical speed standard deviation offset. The conversion method is to multiply the speed fluctuation coefficient by the average speed of the auxiliary motor within the corresponding time window to obtain the actual speed fluctuation amplitude during the abnormal period. This fluctuation amplitude serves as the reference value for compensation, and its direction is determined based on the power characteristics of the auxiliary motor: if the auxiliary motor is underloaded, positive compensation is applied to increase the speed; if it is overloaded, negative compensation is applied to reduce the speed requirement. The final generated dynamic energy efficiency adjustment parameters include the torque compensation value of the main drive motor and the speed compensation value of the auxiliary motor, along with a timestamp. These parameters are encapsulated as data instruction packets and temporarily stored in the sending buffer of the edge node. The entire calculation process is completed before the generation of the next multidimensional state matrix to meet the system's real-time response requirements.

[0070] Example 3: See Figure 4The transmission of dynamic energy efficiency adjustment parameter commands utilizes a real-time communication interface between the edge computing node and the press controller. This interface is based on the Industrial Ethernet protocol, and its transmission cycle is synchronized with the scan cycle of the press control system. The edge node writes the encapsulated adjustment command into a designated data area of ​​shared memory, which the controller reads at the beginning of each control cycle. The command data packet contains three core fields: the torque compensation value of the main drive motor, the speed compensation value of the auxiliary motor, and the command activation timestamp. The timestamp ensures that the controller is strictly synchronized with other motion control signals when executing the command. Hardware-level verification mechanisms are implemented during command transmission, such as a cyclic redundancy check (CRC) code appended to the end of the data packet, to prevent data tampering or loss during transmission.

[0071] The torque compensation applied to the main drive motor is divided into two stages. The first stage involves dynamic limiting. The controller obtains the maximum permissible output torque value specified by the motor manufacturer and compares the absolute value of the torque compensation with this limit. When the absolute value of the compensation exceeds 50% of the limit, the system reassigns the compensation according to the sign of 50% of the limit. The second stage is torque command reconstruction. The controller extracts the original torque command value of the main drive motor for the current work cycle and linearly superimposes the limited torque compensation onto this command value. The superposition process is not a simple arithmetic addition but involves interpolation based on the real-time working phase of the press. During the press slide's upward phase, the compensation is accumulated in five steps over 200 milliseconds; during the slide's downward phase, a three-step accumulation strategy is used to avoid sudden torque changes that could cause mechanical shock. The reconstructed torque command value is sent to the current loop input port of the motor driver.

[0072] The speed compensation is applied to the closed-loop control circuit of the auxiliary motor. The speed regulation module of the controller includes a proportional-integral-derivative (PID) control loop. The compensation is input to the setpoint processing channel of this module. In the standard procedure, the setpoint processing channel first performs a ramp transition processing on the speed setpoint to generate a smooth command, and then superimposes the speed compensation through a first-order low-pass filter onto this command. The cutoff frequency of the low-pass filter is automatically adjusted according to the mechanical time constant of the auxiliary motor rotor to ensure that the introduction of the compensation does not induce mechanical resonance. The superimposed combined speed command is input to the input of the PID controller. The output of the PID controller serves as the setpoint for the current loop, ultimately driving the auxiliary motor to perform the speed regulation task.

[0073] After updating the power allocation strategy, the collaborative performance of the two motors needs to be verified. The edge computing node re-collects the three-phase current signals of the two motors during the update period. The current signal sampling window length is set to an integer multiple of the press's working cycle. Fast Fourier Transform is performed on both sets of current signals to extract the fundamental component and calculate the phase angle. The formula for calculating the phase angle difference is:

[0074] ;

[0075] in: The updated average phase difference angle (degrees). It is the number of sampling points. For the sampling time point, and They represent the first The current phase angles of the main drive motor and auxiliary motor at each sampling point are calculated. This result is compared with a preset cooperative tolerance range. The tolerance range is dynamically set according to the press's operating stage: the upper limit is widened to 8 degrees during the slide acceleration stage and tightened to 3 degrees during the forming and pressurizing stage. When the average phase difference exceeds the tolerance range of the current stage, the system marks a cooperative anomaly.

[0076] The feedback adjustment mechanism is automatically activated after the power strategy is updated. Edge nodes synchronously acquire slider displacement acceleration signals before and after the update. The raw accelerometer data is normalized to eliminate the influence of different time scales. The ratio of the variances of the slider displacement acceleration before and after the update is calculated.

[0077] System defines variance change rate parameter ,in This represents the variance of the slider displacement acceleration after the strategy update. This represents the original variance value before the update. The preset variance change threshold. The value is 0.25.

[0078] When detected At this time, the compensation adjustment range reduction procedure is triggered. The reduction operation adopts a step-by-step iterative strategy: first, the absolute value of the current torque compensation is multiplied by a reduction factor of 0.7, and the absolute value of the speed compensation is multiplied by a reduction factor of 0.6, while keeping the sign of the original compensation direction unchanged. The reduced new compensation is then re-input into the torque command reconstruction and speed closed-loop control loop. Simultaneously, the trigger delay time of the collaborative tolerance verification stage is adjusted to 400 milliseconds to allow the mechanical system sufficient time to respond and adjust.

[0079] The system establishes a closed-loop evaluation mechanism for adjustment results. After each reduction in compensation, two sets of data are re-acquired: updated current phase difference information and slider displacement acceleration signal. The phase difference data is then substituted into the collaborative verification formula for recalculation. If the value still exceeds the tolerance range, a second reduction process is initiated. The second reduction uses a stronger reduction factor, lowering the torque compensation factor to 0.4 and the speed compensation factor to 0.3. A time-domain analysis is performed on the variance of the slider displacement acceleration to detect the presence of high-frequency vibration components. After two consecutive reduction operations... If the value is still greater than the threshold, it is determined to be an abnormal operating condition. The system then stops the automatic adjustment process and issues an equipment maintenance request signal.

[0080] Edge computing nodes record complete adjustment process data. The records include initial adjustment parameter values, reduction ratios for each adjustment, phase difference verification results, variance change rate values, and the final adjustment effect code. The data is stored in binary format, with each adjustment event generating a separate data file. The file header contains information such as a timestamp, press model code, and workpiece batch number. These files are periodically uploaded to the central analysis system for long-term operational mode optimization studies. In the local storage system, detailed data from the latest three adjustment cycles is stored in high-speed solid-state storage for rapid backtracking analysis. When local storage space reaches 90% capacity, the system automatically deletes the oldest 20% of data records to free up space.

[0081] Continuous monitoring of dynamic parameters is performed in the background. Edge nodes maintain a circular monitoring buffer, continuously tracking the real-time values ​​of three key parameters: the updated real-time comprehensive energy efficiency index, the running average of the current slider displacement acceleration variance, and the average phase difference of the dual-motor current. These parameters are sampled at 50-millisecond intervals. The monitoring algorithm calculates the moving standard deviation of the three parameters. When the standard deviation of any parameter exceeds its corresponding threshold for three consecutive sampling periods, the system automatically loads the most recently successfully adjusted record data, resets the torque compensation and speed compensation to those historical values, and backs up the current adjustment parameters before the reset operation. After the reset is complete, the collaborative performance verification process is restarted, forming an autonomous recovery mechanism. All reset events generate detailed operation logs, including reset reason analysis codes and historical parameter matching indices.

[0082] Example 4: See Figure 5 The process of determining the stable energy efficiency range is executed through a timed task on the edge computing node. The system is configured with a monitoring sequence of three consecutive time windows, each with a length set to twice the standard working cycle of the pressure machine. Window advancement adopts a progressive overlapping method, with the start point of a new window and the end point of the previous window spaced one-third of the window length apart. When the task starts, it first loads three sets of real-time comprehensive energy efficiency index raw data for the current time window and the previous two time windows. Each set of data contains the energy efficiency index values ​​of all sampling points within that window period, with a sampling frequency maintained at 100 times per second. After data loading, preprocessing is performed: a moving average filter is used to smooth random fluctuations, with the filter window width set to five percent of the sampling points. The preprocessed data enters the standard deviation calculation module. This module uses an unbiased estimation algorithm to calculate the standard deviation of the energy efficiency index values ​​within each window. The calculation first calculates the arithmetic mean of all energy efficiency values ​​within the window, then calculates the squared deviation of each energy efficiency value from the mean, divides the sum of the squared deviations by the number of data points minus one, and finally takes the square root to obtain the standard deviation value for that window.

[0083] A dynamic maintenance mechanism for the target energy efficiency threshold operates synchronously. The system is linked to the press process database and retrieves historical best energy efficiency records based on the material code and thickness parameters of the currently processed workpiece. If the historical database contains more than ten valid records under the same process parameters, 85% of the highest energy efficiency index value among these records is used as a temporary threshold; if no matching record is found, a preset default threshold is used. The temporary threshold is combined with the standard deviation calculated for the current window for analysis: when the standard deviation of three consecutive windows is less than 5% of the target energy efficiency threshold, and the arithmetic mean of the energy efficiency index within each window is higher than 90% of the threshold value, the system generates an energy efficiency stable state identifier. This identifier is written to the control status register, simultaneously triggering a control strategy switch—transferring the dynamic adjustment mode to a stable maintenance mode.

[0084] The continuous monitoring process remains active under stable conditions. The monitoring frequency is adjusted to four times per minute. The latest three-window energy efficiency index standard deviation is recalculated with each monitoring cycle. When any window standard deviation exceeds 5.5% of the target energy efficiency threshold, the system immediately initiates an anomaly response protocol. The first step of the protocol is to retrieve the dual-motor cooperative load matrix records related to the previous state from the historical database. The search criteria are set as follows: within the most recent 72 hours, the torque fluctuation coefficient error does not exceed 10% of the current value, and the speed fluctuation coefficient error is controlled within 8% of the current value. The search results are then displayed in reverse chronological order, showing all matching records.

[0085] The historical data analysis module processes the search results. For each matching record, it extracts the associated slider displacement acceleration data, which is already tagged by time period in the original storage. Acceleration data is grouped: records with the same process parameters are grouped together. Within each group, the variance of acceleration for all records within the equivalent operating time period is calculated, and a weighted average of the variances for that group is obtained. A time decay strategy is used for weight allocation; the weight for the current moment is one, and the weight decreases by 0.05 for each hour in advance. The search results for the data matching process are shown in Table 1.

[0086] Table 1: Retrieval Results of the Data Matching Process

[0087]

[0088] The current operating condition is determined based on automatic grouping of process parameters. The system filters a subset of historical records whose process parameters perfectly match the current processing task. Within this subset, a double-weighted average is calculated for the mean of the acceleration variance: the weight is based on the reciprocal of the time interval between the historical record's generation time and the current time; records with an interval exceeding six hours have their weight reduced to 0.5. The calculated weighted average variance is used as a historical benchmark reference value.

[0089] The anomaly decision module synchronously collects real-time data of the current slider displacement acceleration. It calculates the current acceleration variance using a window of equal time length. The deviation rate is obtained by dividing the absolute difference between this value and the aforementioned historical reference value by the reference value. When the deviation rate exceeds a 15% threshold, a reset instruction code is written to the control register. The reset operation first backs up all current dynamic energy efficiency adjustment parameters to an isolated storage area, and then locates the adjustment parameter combination corresponding to the optimal energy efficiency moment based on the matching history. If multiple high-efficiency points exist in the matching record, the parameter set with the smallest acceleration variance is selected first. The parameter initialization process is executed in steps: the torque compensation amount transitions to the historical value within 0.5 seconds using a three-segment ramp function, and the speed compensation amount gradually changes to the target value within 0.8 seconds using five-segment piecewise linear interpolation. The reset signal simultaneously triggers the power metering module to start, recording the power change waveform before and after the initialization operation.

[0090] Following parameter initialization, system behavior enters a special monitoring cycle. During the first five working cycles, energy efficiency index calculations are performed at double the frequency, and the construction window length of the standard collaborative load state matrix is ​​shortened to 60% of its normal value. Real-time integrated energy efficiency index data is marked as being in the initialization recovery state; its determination results are used only for local policy control and do not participate in long-term database quality index statistics. Each initialization event records a complete trajectory in the event log: including the trigger deviation rate value, the historical record identifier used, and snapshots of key parameters before and after initialization. Log entries include a 13-digit timestamp and the device hardware serial number encoding.

[0091] The local storage system implements lifecycle management for historical load matrices. Load matrices generated at the current moment are marked as priority protection, retaining at least twelve hours of original data. Matrix data exceeding this time limit is losslessly compressed and transferred to secondary storage. The compressed package header includes matrix dimension descriptors and eigenvalue statistical summaries. The system performs storage space maintenance tasks daily at midnight: deleting ordinary matrix data older than seven days, but retaining all matrix records marked as initialization reference points. These critical records are permanently stored in non-volatile memory in their original data format and simultaneously uploaded to a cloud disaster recovery server, forming dual storage protection.

[0092] Example 5: When the edge computing node performs the core energy efficiency optimization task, it simultaneously initiates the data archiving and transmission process. Dynamic energy efficiency adjustment parameters and corresponding operating parameters for the acquisition cycle form a structured data packet. The torque compensation values ​​of the main drive motor and the speed compensation values ​​of the auxiliary motor are stored in four-byte floating-point format. The operating parameter set includes the raw sampling array of the torque sensor, the dual-current phase difference sequence, and the raw data stream of the accelerometer within the adjustment cycle. Each data type is accompanied by a timestamp tag accurate to microseconds. The data packet organization adopts a hierarchical encapsulation architecture: the base layer is a binary data block, the metadata layer contains sensor calibration coefficients and acquisition frequency information, and the application layer records process parameter codes and equipment serial numbers. After encapsulation, the data enters the compression stage. The compression engine uses a dictionary-based encoding algorithm to establish a dynamic symbol table for recurring patterns. The encoding process prioritizes large-capacity floating-point arrays, identifying intervals in continuous sampled values ​​that can be expressed by run-length encoding. The maximum size of the compressed data block is limited to 75% of the storage buffer capacity.

[0093] Data encryption is performed immediately after compression. The encryption module loads the master key stored in the hardware security area and generates a session encryption key through a key derivation function. An encryption mode with authentication functionality is employed, which simultaneously generates ciphertext and integrity check tags. The encryption operation uses 128-bit data blocks as processing units, with each block undergoing sixteen rounds of substitution-permutation transformations. In each round, the data block and the round key undergo logical operations, changing data characteristics through non-linear byte substitution, scrambling spatial positions through row and column shifting operations, and finally performing column mixing transformations. An eight-byte message authentication code is appended to the complete ciphertext output. A version identifier and initialization vector are added to the header of the encrypted data packet, and a cyclic redundancy check sequence is embedded at the tail. The entire encryption process is completed within the cryptographic coprocessor of the edge node, and the key material only exists briefly in the secure enclave memory.

[0094] The Industrial Internet of Things (IIoT) transmission protocol adopts a standardized communication framework. The protocol stack includes transport layer security session guarantees and a two-way digital certificate authentication mechanism to verify the legitimate identities of cloud servers and edge nodes. The physical link supports dual channels of wired industrial Ethernet and cellular networks, with the transmission manager dynamically selecting the primary channel based on current network quality indicators. A data fragmentation mechanism splits large data packets into fragments conforming to the Maximum Transmission Unit (MTU) standard, with each fragment appended with a fragment sequence number and a total packet identifier. The transmission controller maintains a dynamic fragment size table, automatically reducing the fragment size when three consecutive fragments are lost. The basic transmission cycle is set to a fixed interval of five minutes or an integer multiple of five work cycles completed by the stress tester, whichever comes first. Each transmission event triggers the local storage system to create a transmission log, recording the data packet hash value, the destination server address, and the actual transmission time.

[0095] The cloud server receiver is configured with a load balancing mechanism. The ingress gateway parses the packet header information and verifies the validity of the message authentication code. If the integrity verification fails, the gateway immediately sends a retransmission request to the edge node, and if more than three failures occur within three minutes, the node's transmission channel is frozen. The decryption engine uses the corresponding key in the cloud key repository to reverse the decryption process and restore the original data packet. The decompression service adopts a stream processing mode, restoring the original data structure by rebuilding the encoding dictionary. The decompressed data packet is split into two logical parts: dynamic energy efficiency adjustment parameters are imported into the short-term analysis buffer, and the complete operating parameter sequence is written to the time series database.

[0096] The time-series database employs a clustered deployment architecture. Data partitioning is distributed across different storage nodes based on the hash value of the first three characters of the device serial number. Each record stores the following elements: an arrival time tag accurate to milliseconds, unparsed binary payload data, and a packet hash fingerprint. The database automatically creates multi-level indexes: the primary index is organized according to the device serial number and time range, while the secondary index marks process parameter codes and key feature value ranges. The data retention strategy employs a two-tiered storage mechanism: recent data is stored on a solid-state drive array, while historical data older than thirty days is transferred to cold storage media. Upon importing compressed parameter data, a tag extraction operation is triggered to identify torque compensation and speed compensation values ​​in the data packets and establish a mapping relationship with metadata in the device registry.

[0097] The long-term performance analysis module initiates an asynchronous data processing pipeline. The analysis engine scans for new data records every twenty minutes. The data processing phase separates three sets of features: extracting the time series of torque compensation for three consecutive transmission cycles, calculating its average and standard deviation; analyzing the gradient of auxiliary motor speed compensation; and associating it with maintenance status codes in the equipment operation log. The feature set is input into the performance evolution evaluation algorithm, which includes a sliding time window comparison mechanism. The window width is initialized to a twelve-hour length, sliding four hours at a time. Under the same process parameters, the difference between the average compensation values ​​of the current window and historical windows is compared. If the difference exceeds a preset threshold, a compensation offset alarm event is generated. The seasonal decomposition algorithm partitions the annual data by quarter, detecting whether the compensation parameters exhibit a periodic fluctuation pattern related to temperature. The anomaly pattern recognizer runs the isolated forest algorithm to mark anomalous abrupt changes in the compensation trajectory.

[0098] A separate feedback channel for cloud-based analytics results is established. The feedback data packet contains a unique identifier for the edge node, a time window locator, and parameter adjustment suggestions. These suggestions may include corrected weighting coefficients for the energy efficiency assessment model and updated proportional factors in the compensation calculation. The transmission controller prioritizes using currently idle data channels on the edge node; if the encrypted transmission channel is unavailable, it switches to a low-rate control channel. The edge node receiver is configured with a dedicated message parser to separate valid instruction fields from the feedback packet. Parameter updates are performed during device idle periods: the loading of weighting coefficients for the dynamic energy efficiency assessment model employs a checksum protection mechanism, calculating the checksum twice before and after the update file transfer; the compensation proportional factor is updated via a double-buffer switching mechanism, with older configurations temporarily retained as rollback points. After each successful update, the node sends an application confirmation signal to the cloud, carrying a snapshot of the parameters before the update and the initial state parameters after execution. Unconfirmed feedback instructions will be automatically resent after eighteen hours, and after five consecutive failures, they are marked as invalid.

[0099] The archiving system establishes a closed-loop audit trail mechanism. At the start of transmission, edge nodes generate a transmission event serial number, which is a composite code using year, month, day, hour, minute, second, and device serial number. This serial number is transmitted throughout the data packet's journey to the cloud database for storage. At each critical stage of data processing—after compression, encryption, before transmission, and after cloud decryption—the system records the processor identifier and processing timestamp of the data packet corresponding to that serial number. The complete transmission link's tracking log is independently stored in the cloud audit database, and each transmission link can be reconstructed using the serial number as an index. Edge nodes locally retain detailed transmission logs for the most recent 72 hours; expired logs are compressed with a hash digest and uploaded for archiving.

[0100] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing the energy efficiency of a dual-motor servo press based on edge computing, characterized in that, include: The operating parameters of the main drive motor and the auxiliary motor are collected in real time by the edge computing node deployed on the dual-motor servo press body; The dual-motor collaborative load state matrix is ​​dynamically constructed based on operating parameters, and the real-time comprehensive energy efficiency index is calculated through the energy efficiency evaluation model in the edge computing node. Based on the deviation between the real-time comprehensive energy efficiency index and the target energy efficiency threshold, the dynamic energy efficiency adjustment parameters of the main drive motor and the auxiliary motor are determined. The edge computing node sends dynamic energy efficiency adjustment parameter commands to the controller of the dual-motor servo press, and synchronously updates the power allocation strategy of the main drive motor and the auxiliary motor. The system continuously monitors the real-time comprehensive energy efficiency index and adjusts the dynamic energy efficiency regulation parameters until the dual-motor coordinated load state reaches the preset energy efficiency stability range. The operating parameters include the real-time output torque of the main drive motor, the instantaneous speed of the auxiliary motor, the current phase difference when the two motors work together, and the displacement acceleration of the press slide. The calculation process of the real-time comprehensive energy efficiency index includes: A multidimensional state matrix is ​​input into the energy efficiency assessment model, and the load characteristics of the main drive motor and the auxiliary motor are separated by a spatiotemporal feature extraction network. By integrating load characteristics with preset motor power loss baseline values, dynamic energy efficiency parameters are generated for dual-motor collaborative operation. The dynamic energy efficiency parameters are normalized and weighted to output a real-time comprehensive energy efficiency index in the range of 0-1.

2. The energy efficiency optimization method for a dual-motor servo press based on edge computing according to claim 1, characterized in that, The dynamically constructed dual-motor cooperative load state matrix includes: Extract the torque fluctuation coefficient of the main drive motor and the speed fluctuation coefficient of the auxiliary motor within a preset time window; Calculate the coupling influence factor between the phase difference of the dual motor currents and the acceleration of the slider displacement; The torque fluctuation coefficient, speed fluctuation coefficient, and coupling influence factors are integrated into a multi-dimensional state matrix according to the time series.

3. The energy efficiency optimization method for a dual-motor servo press based on edge computing according to claim 1, characterized in that, The determination of the dynamic energy efficiency adjustment parameters for the main drive motor and the auxiliary motor includes: Establish an energy efficiency change curve with real-time comprehensive energy efficiency index as the vertical axis and timestamp as the horizontal axis; Identify the periods in the energy efficiency change curve where the energy efficiency is continuously below the target energy efficiency threshold, and extract the maximum torque fluctuation coefficient and minimum speed fluctuation coefficient corresponding to those periods. Based on the ratio between the maximum torque fluctuation coefficient and the standard torque threshold, the torque compensation amount of the main drive motor is generated. The speed compensation amount for the auxiliary motor is generated based on the difference between the minimum speed fluctuation coefficient and the reference speed.

4. The energy efficiency optimization method for a dual-motor servo press based on edge computing according to claim 3, characterized in that, The update of the power allocation strategy includes: The torque compensation is added to the current output torque command value of the main drive motor; The speed compensation is embedded in the closed-loop control feedback loop of the auxiliary motor; The updated dual-motor current phase difference is verified using edge computing nodes to ensure it meets the preset collaborative tolerance range.

5. The energy efficiency optimization method for a dual-motor servo press based on edge computing according to claim 4, characterized in that, The feedback adjustment process includes: After the power allocation strategy is updated, the slider displacement acceleration during the update period is re-acquired; Calculate the rate of change of variance of the slider displacement acceleration before and after the update; If the variance change rate exceeds the preset threshold, the adjustment range of torque compensation and speed compensation will be reduced proportionally.

6. The energy efficiency optimization method for a dual-motor servo press based on edge computing according to claim 5, characterized in that, The determination of the energy efficiency stability range includes: Continuously monitor the real-time comprehensive energy efficiency index for three consecutive time windows; When the standard deviation of the real-time comprehensive energy efficiency index in the three time windows is less than the preset tolerance value and is higher than 90% of the target energy efficiency threshold, it is determined that the energy efficiency has entered the stable range.

7. The energy efficiency optimization method for a dual-motor servo press based on edge computing according to claim 1, characterized in that, When performing energy efficiency optimization, the edge computing node simultaneously compresses the dynamic energy efficiency adjustment parameters and corresponding operating parameters into encrypted data packets, which are then transmitted to the cloud server via the Industrial Internet of Things protocol for long-term performance analysis.

8. The energy efficiency optimization method for a dual-motor servo press based on edge computing according to claim 7, characterized in that, The continuous monitoring process for the energy efficiency stability range includes: When the standard deviation of the real-time comprehensive energy efficiency index exceeds the preset tolerance value, the preceding data of the current dual-motor collaborative load state matrix is ​​retrieved from the historical database of the edge computing node. Extract historical records from the preceding data that match the current torque fluctuation coefficient and speed fluctuation coefficient, and calculate the mean variance of the slider displacement acceleration corresponding to the historical records; The current variance of slider displacement acceleration is compared with the historical mean variance. If the deviation exceeds the threshold, the initialization and reset of the dynamic energy efficiency adjustment parameters are triggered.

Images