Electric feeder positioning detection system and method

Abstract

The invention discloses an electric feeder positioning detection system comprising an initialization control module used for collecting signals, detecting stability, establishing a system reference position and sending a control instruction; the instruction processing module is used for converting the control instruction into a motor control parameter and providing a motion planning basis; the motor driving module is used for receiving motor control parameters and performing dynamic position tracking in real time; the closed-loop control module is used for correcting step loss or overshoot in real time based on position feedback; and the automatic learning module provides fine tuning parameters according to environmental changes. The system further comprises a health monitoring module, and the health monitoring module comprises a calculation error statistical unit and an early warning unit. Through multi-sensor information fusion and AI-driven closed-loop control, the system automatically returns to zero after being electrified, errors are eliminated, and the robustness in a complex industrial environment is improved. And the strict requirements of a modern SMT production line on high precision, high efficiency, high reliability and intelligence can be met.

Description

Technical Field

[0001] This invention relates to the field of positioning detection, and specifically to an electric feeder positioning detection system and method. Background Technology

[0002] With the continuous expansion of demand in the electronic information industry and the continuous improvement of electronic information technology infrastructure, more and more automated production processes such as electronic manufacturing, assembly line production, and unmanned assembly are placing higher demands on the performance indicators of industrial equipment, such as precision and speed.

[0003] In SMT (Surface Mount Technology) production lines, the positioning accuracy of feeders directly affects the success rate of component pick-up and placement quality. Current mainstream electric feeders and their positioning detection systems suffer from insufficient detection accuracy and resolution, failing to meet the demands of high-precision, miniaturized component placement. These systems also exhibit poor anti-interference capabilities, inadequate stability and reliability, and cannot match the rapid start-up and precise positioning requirements of high-speed electric feeders. Slow detection feedback in high-speed placement mode becomes a bottleneck, forcing the entire production line to slow down, reducing production efficiency. Furthermore, the system is complex, has low integration, and requires significant daily maintenance. Once positioning drift occurs, the recalibration process is time-consuming and labor-intensive, further impacting overall equipment efficiency.

[0004] Existing systems typically only provide simple binary status signals such as "arrived / not arrived," lacking process data and intelligent diagnostic and adaptive capabilities. Current technical solutions have significant shortcomings in terms of accuracy, speed, reliability, ease of use, and intelligence, failing to reliably meet the stringent requirements of modern SMT production lines for high precision, high efficiency, high reliability, and intelligence. Summary of the Invention

[0005] This invention provides an electric feeder positioning and detection system and method to solve the problems mentioned in the background art, where existing systems typically only provide simple binary state signals, lack process data, and lack intelligent diagnostic and adaptive capabilities; thus failing to reliably meet the stringent requirements of modern SMT production lines for high precision, high efficiency, high reliability, and intelligence.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an electric feeder positioning and detection system, comprising: Initialize the control module, acquire signals and detect stability, establish the system reference position, and send control commands; The instruction processing module converts control instructions into motor control parameters, providing a basis for motion planning. The motor drive module receives motor control parameters and performs dynamic position tracking in real time. The closed-loop control module, based on position feedback, corrects lost steps or overshoot in real time; The automatic learning module provides fine-tuning parameters based on changes in the environment.

[0007] It also includes a health monitoring module, which comprises an error statistics unit and an early warning unit. The error statistics unit calculates error statistics. ; ; Let be the sample size, if The system triggers an alert; the alert unit uses lexical relevance to assess the probability of hallucination. , To generate a reference sequence, if An early warning was issued.

[0008] Preferably, the initialization control module includes a signal acquisition unit and an absolute coordinate setting unit, wherein the signal acquisition unit is used to calculate the signal fluctuation variance. ,in For variance, The number of sampling points. For the i-th sample value, For the average value, if If the signal is stable, then the absolute coordinate setting unit is used to construct the absolute coordinates of the current position. ,in , To stabilize the time, .

[0009] Preferably, the instruction processing module includes an instruction parsing unit and a pulse calculation unit, wherein the instruction parsing unit is used to receive target position instructions. Calculate relative displacement The pulse calculation unit is based on the magnetic encoder resolution. Calculate the required number of pulses ,in For gear ratio, Where is the radius of the motor.

[0010] Preferably, the motor drive module includes a motor control unit and a pulse reading unit, wherein the motor control unit is used to output drive signals. ,in , For the current count, This is a proportionality coefficient; the pulse reading unit reads the pulse count in real time. ,like If the motor stops, stop it; otherwise, continue.

[0011] Preferably, the closed-loop control module includes an error calculation unit and a motion adjustment unit, wherein the error calculation unit is used to calculate the position error. ,in Reconstruct the spatial spectrum at the current location using the MoD-DNN model. , As weight, The phase is the calibration angle, which depends on the phase error; the motion adjustment unit outputs... ,in For the integral and differential coefficients, the planned trajectory is optimized using the least squares method. , and train the DNN by fusing synthetic data.

[0012] Preferably, the automatic learning module includes an environmental parameter acquisition unit and a parameter fine-tuning unit, wherein the environmental parameter acquisition unit acquires temperature data. and wear indicators Using the EMDB dataset to train a 3D pose model to estimate mechanical offset The parameter fine-tuning unit automatically adjusts the PID coefficients. , The learning rate is used for gradient descent optimization. , For loss function, The step size.

[0013] This invention also discloses an electric feeder positioning and detection method, based on the aforementioned electric feeder positioning and detection system, comprising the following steps: S1. System initialization and mechanical zeroing: After the system is powered on and started, the controller drives the motor to search for the mechanical reference point, and uses the micro-motion reference point sensor to detect the signal stability to achieve absolute coordinate setting. S2. Command Reception and Pulse Calculation: The controller receives control commands and calculates the required number of pulses using the magnetic encoder. S3. Motor drive and real-time monitoring: Drive the motor forward and continuously read the magnetic encoder pulse signal to form a preliminary closed loop; S4. Closed-loop control and error correction: A PID algorithm based on position feedback is introduced to correct lost steps or overshoot in real time.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention cross-integrates UAV-assisted positioning technology from wireless sensor networks, Model-Driven Deep Neural Networks (MoD-DNN) for angle estimation, and health status monitoring concepts from the field of psychology for system self-diagnosis. Through multi-sensor information fusion and AI-driven closed-loop control, the system automatically zeroes out upon power-on, eliminating errors and improving robustness in complex industrial environments. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0017] This embodiment discloses an electric feeder positioning and detection system, including: Initialize the control module, acquire signals and detect stability, establish the system reference position, and send control commands; The instruction processing module converts control instructions into motor control parameters, providing a basis for motion planning. The motor drive module receives motor control parameters and performs dynamic position tracking in real time. The closed-loop control module, based on position feedback, corrects lost steps or overshoot in real time; The automatic learning module provides fine-tuning parameters based on changes in the environment; The health monitoring module analyzes current and error data to monitor the system's health status in real time. The purpose and function of this step is to shift from reactive to proactive maintenance, reduce downtime, and enhance the system's self-diagnostic capabilities through cross-disciplinary psychological knowledge.

[0018] Based on the above modules, this embodiment discloses an electric feeder positioning and detection method, including: S1. System Initialization and Mechanical Zeroing: After the system is powered on and started, the controller drives the motor to search for the mechanical reference point. The micro-motion reference point sensor detects signal stability, thus achieving absolute coordinate setting. The purpose and function of this step is to establish the system reference position, eliminate historical errors, and ensure the absolute accuracy of subsequent positioning. Specifically, this includes: S11. Signal acquisition and stability detection: The controller drives the motor at a constant speed. Moving forward while continuously acquiring signals from the micro-motion reference point sensor. To determine if a signal has entered a stable low level: calculate the signal fluctuation variance. ,in For variance, The number of sampling points. For the i-th sample value, This is the average value. If... (Threshold θ is set to 0.01V) 2 If the signal is stable, then the signal is stable. S12. Absolute coordinate setting: Once stable, set the current position as the absolute coordinate. ,in , To stabilize the time, (Assuming linear motion). Accumulated errors are eliminated by resetting the encoder count. .

[0019] S2. Command Reception and Pulse Calculation: The controller receives control commands and calculates the required number of pulses using the magnetic encoder. After zeroing, the controller receives feeding commands and calculates the required number of pulses using the high-precision magnetic encoder. The purpose and function of this step is to transform abstract commands into executable motor control parameters, providing a precise basis for motion planning. Specifically, this includes: S21. Command parsing, receiving target location command. Calculate relative displacement The initial path loss factor is estimated by fusing sensor data using an RSS-based method. ,in For received power, For transmission power, For distance, The path index (typically 2-4); S22. Pulse count calculation, based on magnetic encoder resolution. (Pulse / Revolution), calculate the required number of pulses. ,in For gear ratio, Given the motor radius. The Time-of-Availability (ToA) method is introduced to calibrate time synchronization. , To ensure the speed of light, calculation accuracy must be guaranteed.

[0020] S3. Motor Drive and Real-time Monitoring: The motor is driven forward while continuously reading the magnetic encoder pulse signal, forming a preliminary closed loop. The purpose and function of this step is to achieve dynamic position tracking, ensuring high-precision monitoring and immediate response during the feeding process. Specifically, this includes: S31, Motor starts, controller outputs drive signal ,in , For the current count, This is a proportionality coefficient. Monitoring current. To detect abnormalities , For periodicity; S32. Pulse reading and judgment, real-time reading of pulse count. ,like (ε is 0.1 pulse) Stop the motor; otherwise continue. Calculate the position difference using the fused TDoA method. , This is the time difference of arrival.

[0021] S4. Closed-loop control and error correction: Introducing a position feedback-based PID algorithm to correct missed steps or overshoot in real time. The purpose and function of this step is to provide real-time error correction, ensuring stability and accuracy at high speeds to meet the requirements of miniature component placement. Specifically, it includes: S41. Error Calculation: Calculate the position error. ,in Current location. Reconstructing the spatial spectrum using the MoD-DNN model. , As weight, For phase, the calibration angle depends on the phase error; S42, PID Tuning and Motion Planning, PID Output ,in For the integral and differential coefficients, the planned trajectory is optimized using the least squares method. By fusing synthetic data to train DNNs, robustness can be improved.

[0022] In further embodiments, a command parsing and pulse calculation method integrating RSS / ToA / TDoA approaches is presented, focusing on innovative improvements in multi-sensor fusion (RSS / ToA / TDoA) and high-precision positioning. In the command parsing and pulse calculation stage, the system needs to calculate the relative displacement of high-level targets. Mapped to the number of pulses of a magnetic encoder As a direct control quantity for motor drive, it also assists in correcting displacement estimation through multi-sensor fusion (RSS, ToA, TDoA, and other wireless positioning methods). The goal is to overcome the limitations of traditional wireless positioning methods in industrial scenarios (such as low RSS accuracy, complex ToA clock synchronization, and TDoA environmental sensitivity), reliably converting abstract feeding commands into executable pulse commands, and balancing the dual requirements of absolute positioning and relative pulse closed-loop control. Specifically, this includes: S211. Construct a multimodal spatiotemporal feature fusion framework. The system architecture consists of three layers: The underlying data acquisition layer includes a high-precision magnetic encoder (providing pulse increments), an absolute position sensor (providing a drift-free reference), and a wireless positioning module (RSS / ToA / TDoA signal acquisition). The middle layer feature processing layer: drawing on the idea of ​​"feature extraction and registration" in computer vision, maps multi-source data into a unified "spatiotemporal feature map" and dynamically weights it through an attention mechanism; High-level instruction parsing layer: converts the fused position estimate into pulse count. And combined with closed-loop control to complete motor drive.

[0023] S212. Based on the spatio-temporal attention mechanism, a spatio-temporal attention fusion algorithm (STAF) is proposed. The specific steps are as follows: S2121, Data Preprocessing and Feature Extraction: The magnetic encoder provides high-frequency incremental pulse data, denoted as , indicating time Increment of position within; The absolute position sensor provides a low-frequency absolute position reference, denoted as ,in For discrete calibration times; The wireless positioning module provides RSS, ToA, and TDoA signals, which are denoted as follows: ,express Signal strength and path loss factor at each anchor point; This indicates the arrival time; , which represents the differential arrival time.

[0024] S2122, Spatiotemporal Feature Map Construction: Drawing inspiration from image registration in computer vision, this method maps multi-source data into a unified "spatiotemporal feature map". ,in Indicates the time dimension. This represents the spatial dimension (corresponding to different sensors or anchor points). The formula for constructing the feature map is: Symbol explanation: The initial weights of each data source are determined by the sensor reliability and historical noise levels. : The nonlinear mapping function from signal to position, obtained through pre-calibration.

[0025] S2123, Dynamic weighting of attention mechanisms: Drawing inspiration from the attention mechanism in NLP, dynamic weights are calculated for each data source at different time and spatial points, ensuring that higher-confidence data is assigned higher weights. The formula for calculating attention weights is: ; in: Data source exist Attention weights at each location. : Scoring function that measures the relevance of data source features to the context. Context vector, which integrates historical location and environmental interference information. The system features a trainable parameter matrix that learns offline using historical data. Through an attention mechanism, it can dynamically identify environmental interference (such as RSS multipath effects) or sensor anomalies (such as ToA clock drift), automatically reducing the weight of unreliable data to ensure the robustness of the fusion results.

[0026] S2124. Fusion Location Estimation: Calculate the fusion location estimate based on the attention-weighted feature map. : ; This formula integrates multi-source data by weighting according to credibility, and outputs the final location estimate.

[0027] S2125, Pulse Number Calculation and Correction: Estimating the Fusion Position Mapped to target pulse number The calculation formula is: ; in: : This is the difference between the target position and the current position. This represents the physical displacement (unit: micrometer / pulse) corresponding to a single encoder pulse. The correction term, based on historical errors, is dynamically updated through online learning. The formula is: ; Symbol explanation: This is the correction factor, which controls the correction range. This is the length of the historical window used for error smoothing. (Correction item) By introducing the concept of "historical error averaging" and combining it with online learning, it can dynamically adapt to system deviations caused by mechanical wear or environmental changes.

[0028] S213, Industrial Scene Adaptability Optimization S2131, Fixed Anchor Point Topology Design: Deploy fixed UWB anchor points (based on TDoA / phase ranging) in the factory environment, optimizing the point layout to reduce occlusion and multipath interference. Anchor point positions are determined by generating a calibration matrix through offline calibration. Used to correct wireless positioning errors: ; S2132, Hierarchical control logic: Power-on / zeroing phase: Prioritize using absolute sensors (such as absolute magnetic encoders) as the reference; Normal operation: mainly based on magnetic encoder pulses and PID closed loop, with wireless positioning data only used for long-term drift detection; Abnormal Trigger: Enable wireless positioning calibration when encoder and absolute sensor data are inconsistent.

[0029] S2133. Simulation Verification and Adaptive Adjustment: Construct an end-to-end simulation model (including motor dynamics, encoder noise, and wireless signal multipath interference) and generate synthetic data to evaluate the performance of the STAF algorithm in different scenarios. Based on the simulation results, dynamically adjust the attention mechanism parameters and correction coefficients. .

[0030] This embodiment proposes a multimodal fusion framework (STAF) based on a spatiotemporal attention mechanism by introducing innovative ideas from the fields of computer vision and NLP. This framework effectively solves the accuracy and robustness issues of RSS / ToA / TDoA in industrial scenarios. Combined with quantitative pulse calculation and correction formulas, the system can achieve reliable command parsing and motor control with micron-level positioning accuracy, providing a new technical path for high-precision feeding in electric feeder systems.

[0031] The health monitoring module analyzes current and error data to monitor the system's health status in real time. The health monitoring module includes an error statistics calculation unit and an early warning unit. Error statistics unit calculates error statistics ; ; Given the sample size, a psychological stress model is introduced to assess fatigue similar to "occupational burnout." (δ = 0.5 μm) triggers an alert. The alert unit uses lexical relevance to assess the probability of hallucination. , To generate a reference sequence, if An early warning is issued. The integrated dialogue system interface reports the status.

[0032] In many embodiments, step S4. closed-loop control and error correction have been optimized, such as high-precision angle / phase calibration and real-time control based on the fusion of MoD-DNN and PID closed-loop control, specifically including: Optimizing the Multi-Scale MoD-DNN Architecture: To address complex disturbances in industrial environments (such as electromagnetic noise, mechanical vibration, and temperature drift), a multi-scale feature extraction and fusion MoD-DNN architecture is designed. The original MoD-DNN primarily relies on single-scale spatial spectrum reconstruction, while the new design introduces multi-scale convolutional kernels (inspired by multi-scale feature extraction in computer vision, such as the Inception architecture) to extract phase error features at different frequencies and spatial resolutions, enhancing the model's ability to perceive multi-source disturbances. Simultaneously, an attention mechanism is added to dynamically adjust the weights of features at different scales, improving the calibration accuracy for key phase errors.

[0033] The goal of multi-scale MoD-DNN is to reconstruct the spatial spectrum and estimate the phase error from the input signal. The multi-scale feature extraction and fusion process is defined as follows: ; in: This is the feature map after multi-scale fusion, representing the intermediate representation of the spatial spectrum; For scale quantity (e.g.) (corresponding to 1x1, 3x3, and 5x5 convolution kernels); For the first Attention weights at each scale are dynamically calculated by the attention mechanism; For the first A convolution operation at each scale, with parameters as follows: . The input signal matrix is ​​typically the raw signal acquired by a sensor array.

[0034] Attention weight Calculated using the spatial attention module: ; in: For the first Attention scores on multiple scales; It is a multilayer perceptron used to extract attention scores from features; To reduce the feature map dimension, an average pooling operation is used. The final phase error estimate is: ; in: This is the estimated phase error vector; : Fully connected layer, parameters are The fused features are mapped to phase errors.

[0035] Dynamic PID Parameter Adaptation and MoD-DNN (Model-Driven Deep Neural Network) Co-optimization: In the original scheme, the output of MoD-DNN only served as a reference input for PID control, lacking a dynamic co-optimization mechanism. In the new design, the phase / angle error estimation and its uncertainty quantification from MoD-DNN are fed back to the PID controller in real time. A lightweight online learning module dynamically adjusts the PID parameters (such as proportional gain Kp, integral gain Ki, and derivative gain Kd). Furthermore, a feedforward compensation mechanism is introduced, using the phase error distribution predicted by MoD-DNN to generate a feedforward control signal, reducing the PID response delay and improving stability under high-speed operation. The phase error output of MoD-DNN... and its uncertainty The variance estimate (from the model output) is used to dynamically adjust the PID parameters. The PID control law is defined as follows: ; in: For control output (such as motor drive signal); This represents the error between the current position and the target position. The PID parameters are time-dependent and dynamically adjusted. This is the feedforward compensation term, based on the phase error predicted by MoD-DNN.

[0036] The formula for dynamically adjusting PID parameters is: ; ; ; in: These are the initial PID parameters, serving as the baseline values. To adjust the step size hyperparameter and control the sensitivity of parameter updates; This is a context-aware function implemented using a lightweight neural network, with the phase error as input. Uncertainty and historical context (An error sequence over a number of past moments); A historical context vector, representing the past. Error trend at each moment.

[0037] The feedforward compensation term is calculated as follows: ; in: This is the feedforward gain coefficient; This is a feedforward mapping function that maps the phase error into a control compensation signal.

[0038] Real-time performance and low latency optimization: To address the issue that the inference latency of the original MoD-DNN scheme may not meet the requirements for high-speed closed-loop control, a hybrid inference framework is designed. Initially, a lightweight CNN is used for fast coarse estimation, followed by refinement using a sparse conjugate gradient (SCG) algorithm with limited iterations, ensuring that the total inference time is controlled within milliseconds to meet the real-time requirements of embedded hardware. To meet the real-time requirements, the hybrid inference time constraints are designed as follows: ; in: Total reasoning time; A rough time estimate for a lightweight CNN; For sparse conjugate gradient refinement time (limiting the number of iterations); This is the real-time time threshold required by the system (e.g., 1ms).

[0039] Environmental Adaptation and Long-Term Stability Assurance: To address system distribution drift caused by temperature variations and mechanical wear, a domain-adaptive online fine-tuning mechanism is designed. During the initial operation phase, a pre-trained model (based on synthetic data) provides initial calibration capabilities. Subsequently, online fine-tuning is performed using a small amount of real-world operating data to dynamically update the MoD-DNN parameters, resisting performance degradation over long-term operation. Simultaneously, a health status monitoring module is introduced to predict potential faults and trigger parameter self-calibration by analyzing data such as current and error distribution.

[0040] Drawing inspiration from multi-scale feature extraction and attention mechanisms in computer vision (such as the GoogleNet / Inception architecture and Self-Attention in Transformer), this approach is applied to the spatial spectrum reconstruction and phase calibration tasks of MoD-DNN. The specific innovative ideas are as follows: Multi-scale feature extraction is applied to spatial spectrum reconstruction: In computer vision, multi-scale feature extraction is widely used to process image features at different resolutions to capture local and global information. Inspired by this idea, we designed MoD-DNN as a multi-scale architecture, using convolutional kernels of different sizes (e.g., 1x1, 3x3, 5x5) to extract low-frequency (global phase shift) and high-frequency (local error) features from the spatial spectrum in parallel. Subsequently, a feature fusion layer integrates the multi-scale information, improving the model's ability to model complex phase distortions. Compared to single-scale feature extraction, this method significantly improves robustness in industrial noise environments.

[0041] Attention mechanisms enhance the perception of critical errors: In visual tasks, attention mechanisms are used to focus on key regions in an image. Similarly, we introduce a SpatialAttentionModule in MoD-DNN to dynamically adjust the weights of different regions (corresponding to different angles / phases) in the spatial spectrum, prioritizing the calibration of error components that have the greatest impact on positioning accuracy. This mechanism is particularly suitable for uneven signal interference distribution in industrial environments, effectively improving calibration efficiency.

[0042] Deep Integration with PID Control: In the control field, traditional PID parameter tuning often relies on manual experience or offline optimization. Drawing inspiration from context-aware modeling in Natural Language Processing (NLP) (such as the context dependency of Transformers), we design a context-aware PID parameter tuner. This tuner dynamically generates PID parameters by taking the phase error distribution output by the MoD-DNN and its time-series context (error trends over several past moments) as input. This approach transforms PID control from a static rule-driven process into an adaptive control based on the environment and historical states, significantly improving stability under dynamic disturbances.

[0043] This solution borrows multi-scale feature extraction and attention mechanisms from computer vision across disciplines and applies them to spatial spectrum reconstruction and phase calibration, significantly improving the robustness of MoD-DNN in complex industrial environments. Simultaneously, the context-aware PID parameter tuner design breaks through the static parameter tuning limitations of traditional control theory, achieving deep closed-loop synergy between perception and control.

[0044] Multi-scale MoD-DNN and hybrid inference frameworks can be deployed on edge computing devices (such as FPGAs or embedded GPUs), reducing computational overhead through model pruning and quantization. The online PID parameter adjustment module uses a lightweight network to ensure low-latency operation. The domain-adaptive fine-tuning mechanism can achieve performance optimization with a small amount of real data, making it suitable for industrial field deployment. Through multi-scale MoD-DNN architecture optimization, dynamic PID parameter adaptation, low-latency hybrid inference framework, and environmental adaptation mechanism, the performance of the high-precision electric feeder positioning system is comprehensively improved. Cross-domain innovative ideas and quantitative formula design ensure the scientific validity and engineering feasibility of the solution, providing a solid guarantee for micron-level positioning accuracy and long-term stability.

[0045] In some embodiments, an automatic calibration and learning module is also included. This module enables the system to learn and automatically fine-tune parameters according to environmental changes. This ensures long-term stability and accuracy, adapts to temperature changes and wear, and achieves self-adaptation through AI learning. Specifically, it includes: Environmental parameter acquisition, temperature acquisition and wear indicators Using the EMDB dataset to train a 3D pose model to estimate mechanical offset . Parameter fine-tuning, automatic adjustment of PID coefficients , With a learning rate of 0.01, gradient descent optimization is used. , For loss function, The step size is defined. An error correction prompt is introduced to calibrate sentence-level output. To verify the effectiveness and applicability of the solution, this embodiment designs a series of experiments covering system initialization and zeroing, positioning accuracy testing, multi-sensor fusion effect, PID closed-loop control performance, system health monitoring and early warning capabilities, and adaptive calibration function. The experiments are conducted in a simulated SMT (Surface Mount Technology) production line environment, aiming to evaluate the system's performance at micron-level positioning accuracy and compare it with other traditional methods. Experimental data comes from the following three parts: Real-world industrial environment data: collected from the electric feeder system of a semiconductor chip mounting production line, including motor drive current. Magnetic encoder pulse counting Micro-motion reference point sensor signal and environmental parameters (such as temperature) ,humidity Data acquisition lasted for 7 consecutive days, with a sampling frequency of 1000Hz, totaling approximately 605 million data points. The acquisition equipment included a high-precision magnetic encoder (resolution...). Pulse / revolution), micro switch sensor (sensitivity 0.01V), and current sensor (accuracy 0.1mA).

[0046] Wireless positioning signal data: RSS (Received Signal Strength), ToA (Time of Arrival), and TDoA (Time Difference of Arrival) data were collected by deploying UWB (Ultra-Wideband) devices. Four fixed anchor points were set up in the experimental environment, covering a range of 5m × 5m, with a sampling frequency of 100Hz, collecting approximately 1 million sets of signal data. This signal data was used for multi-sensor fusion positioning calibration.

[0047] Synthetic Data: Synthetic datasets are generated using the TOPO-DataGen tool to simulate abnormal conditions in industrial environments, such as noise, motor step loss, and mechanical wear. The synthetic dataset contains 5000 scenarios, each with 10000 time steps of simulated signals (such as position errors). Phase shift ), used for training and validating MoD-DNN models.

[0048] Before execution, the data must first be preprocessed, specifically including: The collected real data underwent noise reduction processing. A moving average filter (window size 10) was used to smooth signal fluctuations and remove high-frequency noise. Outliers (such as current spikes exceeding 3 standard deviations) were marked and removed, accounting for approximately 0.5%. Data from different sources (such as RSS, current, and pulse count) are standardized into a distribution with a mean of 0 and a variance of 1, which facilitates multi-sensor fusion and model training. By adding Gaussian noise (mean 0, variance range 0.01-0.1) and random offsets, the diversity of synthetic data is enhanced, thereby improving the robustness of the MoD-DNN model.

[0049] S1. System initialization and mechanical zeroing: After the system is powered on, the controller operates at a constant speed. Drive motor to continuously acquire signals from micro-motion reference point sensor. Calculate the variance of signal fluctuations. ,in Number of sampling points, threshold .when When the signal is stable, record the stable moment. and set absolute coordinates ; Intermediate data: The average time required for the signal to stabilize in the experiment is... ,variance The decrease from the initial value of 0.05 to 0.009 indicates that the signal stabilization detection algorithm is effective. The average zeroing error (deviation from the actual mechanical accuracy) is... The standard deviation is .

[0050] S2. Command Reception and Pulse Calculation: The controller receives control commands and calculates the required number of pulses using the magnetic encoder; it also receives target position commands. Calculate relative displacement Estimate path loss factor by combining RSS data. ,in According to the magnetic encoder resolution Pulse / Revolution and Gear Ratio Calculate the number of pulses The motor radius ; Intermediate data: Calculated target pulse number Pulse, RSS-assisted correction path loss factor The position estimation error starts from the initial Reduce to .

[0051] S3. Motor drive and real-time monitoring: Drives the motor forward while continuously reading the magnetic encoder pulse signal to form a preliminary closed loop.

[0052] S4. Closed-loop control and error correction: Introducing a PID algorithm based on position feedback to correct missed steps or overshoot in real time. Position error is calculated in real time. Reconstructing the spatial spectrum using the MoD-DNN model and output phase error The PID control output is... The initial parameters are , , Dynamic parameter adjustment is based on the MoD-DNN output, updating the step size. ; Intermediate data: Position error From the initial state under closed-loop control convergence to The convergence time is Phase error After calibration from Reduce to .

[0053] This experiment also includes: a health monitoring and early warning experiment. Operation process: Calculate error statistics , ,in Set an error variance threshold. ,like This triggered an alert.

[0054] Intermediate data: Variance of error during normal operation In simulated mechanical wear scenarios The alert was successfully triggered. The alert response time is [time]. .

[0055] The experimental results are as follows: Positioning accuracy: The average positioning error of the system in micro-component placement tasks is... The standard deviation is This meets the micron-level precision requirements; Closed-loop control performance: After combining PID closed-loop control with MoD-DNN calibration, the motor step loss rate decreased from 5.2% to 0.3%, and the overshoot decreased from... Reduce to ; Health monitoring and early warning: The system's early warning accuracy rate in simulated fault scenarios is 98.5%, the false alarm rate is 1.2%, and the average early warning lead time is [missing information]. ; Adaptive calibration: under temperature changes Under these conditions, the system automatically fine-tunes the PID parameters, and the positioning error only increases. It exhibits good environmental adaptability.

[0056] The table below shows a performance comparison between this system and traditional methods: This method significantly outperforms traditional methods in terms of positioning accuracy, step loss rate, overshoot, and environmental adaptability, especially under micrometer-level accuracy requirements.

[0057] To analyze the contribution of each module to system performance, the following ablation experiments were conducted: Without MoD-DNN calibration: Removing the MoD-DNN phase calibration module and relying solely on PID control increases the positioning error. The step loss rate rose to 1.5%; Without multi-sensor fusion: using only magnetic encoder data, removing RSS / ToA / TDoA auxiliary positioning, the positioning error increases to [missing information]. Environmental adaptability decreases (error increment is) ); No dynamic PID adjustment: Using fixed PID parameters, the positioning error is... Convergence time from Increase to The ablation test results are shown in the table below: Conclusion: MoD-DNN calibration, multi-sensor fusion, and dynamic PID adjustment all make significant contributions to system performance and are indispensable.

[0058] The above experiments verified the superiority of this embodiment in terms of micron-level positioning accuracy, closed-loop control performance, health monitoring, and adaptive calibration. Experimental results show that this method significantly outperforms traditional open-loop control and single PID control methods, especially in terms of robustness and stability in complex industrial environments. Ablation experiments further revealed the necessity of each module, providing guidance for subsequent optimization. Future work will focus on testing the system in more extreme environments (such as high temperature and high vibration) and further optimizing computational resources.

[0059] Within the technical scope disclosed in this invention, any variations or substitutions that can be easily conceived should be included within the protection scope of this invention. Therefore, the protection scope of this invention should be determined by the scope of the claims.

Claims

1. An electric feeder positioning and detection system, characterized in that, include: Initialize the control module, acquire signals and detect stability, establish the system reference position, and send control commands; The instruction processing module converts control instructions into motor control parameters, providing a basis for motion planning. The motor drive module receives motor control parameters and performs dynamic position tracking in real time. The closed-loop control module, based on position feedback, corrects lost steps or overshoot in real time; The automatic learning module provides fine-tuning parameters based on changes in the environment.

2. The electric feeder positioning and detection system according to claim 1, characterized in that, It also includes a health monitoring module, which comprises an error statistics unit and an early warning unit. The error statistics unit calculates error statistics. ; ; Let be the sample size, if The system triggers an alert; the alert unit uses lexical relevance to assess the probability of hallucination. , To generate a reference sequence, if An early warning was issued.

3. The electric feeder positioning and detection system according to claim 1 or 2, characterized in that, The initialization control module includes a signal acquisition unit and an absolute coordinate setting unit. The signal acquisition unit is used to calculate the signal fluctuation variance. ,in For variance, The number of sampling points. For the i-th sample value, For the average value, if If the signal is stable, then the absolute coordinate setting unit is used to construct the absolute coordinates of the current position. ,in , To stabilize the time, .

4. The electric feeder positioning and detection system according to claim 3, characterized in that, The instruction processing module includes an instruction parsing unit and a pulse calculation unit. The instruction parsing unit is used to receive target position instructions. Calculate relative displacement The pulse calculation unit is based on the magnetic encoder resolution. Calculate the required number of pulses ,in For gear ratio, Where is the radius of the motor.

5. The electric feeder positioning and detection system according to claim 4, characterized in that, The motor drive module includes a motor control unit and a pulse reading unit. The motor control unit is used to output drive signals. ,in , For the current count, This is a proportionality coefficient; the pulse reading unit reads the pulse count in real time. ,like If the motor stops, stop it; otherwise, continue.

6. The electric feeder positioning and detection system according to claim 5, characterized in that, The closed-loop control module includes an error calculation unit and a motion adjustment unit. The error calculation unit is used to calculate the position error. ,in Reconstruct the spatial spectrum at the current location using the MoD-DNN model. , As weight, The phase is the calibration angle, which depends on the phase error; the motion adjustment unit outputs... ,in For the integral and differential coefficients, the planned trajectory is optimized using the least squares method. , and train the DNN by fusing synthetic data.

7. The electric feeder positioning and detection system according to claim 6, characterized in that, The automatic learning module includes an environmental parameter acquisition unit and a parameter fine-tuning unit. The environmental parameter acquisition unit acquires temperature data. and wear indicators Using the EMDB dataset to train a 3D pose model to estimate mechanical offset The parameter fine-tuning unit automatically adjusts the PID coefficients. , The learning rate is used for gradient descent optimization. , For loss function, The step size.

8. A method for positioning and detecting an electric feeder, based on the electric feeder positioning and detection system according to any one of claims 7, characterized in that, Includes the following steps: S1. System initialization and mechanical zeroing: After the system is powered on and started, the controller drives the motor to search for the mechanical reference point, and uses the micro-motion reference point sensor to detect the signal stability to achieve absolute coordinate setting. S2. Command Reception and Pulse Calculation: The controller receives control commands and calculates the required number of pulses using the magnetic encoder. S3. Motor drive and real-time monitoring: Drive the motor forward and continuously read the magnetic encoder pulse signal to form a preliminary closed loop; S4. Closed-loop control and error correction: A PID algorithm based on position feedback is introduced to correct lost steps or overshoot in real time.

Images