Calibration method and device for flexible touch sensor

Abstract

The invention discloses a flexible tactile sensor calibration method and device, and the method comprises the steps: obtaining response data of a flexible tactile sensor array, and carrying out the feature extraction and data preprocessing of the response data; on the basis of the standardized training data, a calibration model is generated through training of a convolutional neural network and a long-short-term memory network; parameters of the environment where the sensor is located are collected, and a Bayesian optimization algorithm is used for self-adaptive adjustment; the sensor is conveyed to a multi-station parallel calibration unit for calibration; and carrying out performance test on the calibrated sensor and generating a unique identification code for quality tracing. According to the invention, the technical problems of strong environment dependence, long calibration time, low efficiency and the like in the calibration process of the existing flexible touch sensor are solved, rapid, high-precision and environment-adaptive flexible touch sensor calibration is realized, the calibration time is shortened from several hours to less than 60 seconds, the calibration error is controlled within + / -2%, and the large-scale production requirement is met.

Description

Technical Field

[0001] The present invention relates to the field of flexible electronic technology, and in particular to a calibration method and device for a flexible tactile sensor. Background Art

[0002] Flexible tactile sensors, core components in emerging fields such as wearable devices, soft robotics, and human-machine interaction systems, are lightweight, flexible, and highly sensitive. They hold broad application prospects in healthcare, industrial automation, and consumer electronics. Currently, flexible tactile sensors are primarily designed based on different operating principles, such as resistive, capacitive, and piezoelectric. They typically consist of a multilayer composite structure composed of conductive, substrate, and functional materials.

[0003] Traditional flexible tactile sensor production technologies primarily rely on methods such as screen printing, inkjet printing, and micro-nanofabrication to fabricate sensing elements. For example, screen printing creates a conductive pattern by transferring conductive ink onto a flexible substrate through a screen, while micro-nanofabrication uses processes such as photolithography and etching to construct microstructures on the flexible substrate, achieving specific tactile sensing functions.

[0004] Existing mass production technology for flexible tactile sensor arrays primarily utilizes a roll-to-roll continuous printing process, combined with automated assembly and packaging systems, and sensor performance testing and calibration at the end of the production line. This technology utilizes a fully automated production line to print a conductive pattern onto a flexible substrate, then laminate and assemble the functional materials, before finally performing cutting, packaging, and performance testing, forming a complete production process.

[0005] However, existing technologies have significant shortcomings in sensor calibration. These include: the calibration process requires a strict constant temperature and humidity environment, placing high demands on production conditions; long calibration times, severely restricting production efficiency; reliance on fixed environmental conditions for calibration parameters, resulting in poor sensor performance consistency in diverse and changing environments; and the bulky, fixed design of the calibration device, lacking flexibility and making it difficult to adapt to large-scale mass production. These issues directly impact the production capacity, yield, and performance stability of flexible tactile sensors, limiting their widespread application. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention provides a calibration method and device for a flexible tactile sensor, which realizes fast, high-precision, and environment-adaptive calibration of the flexible tactile sensor to meet the needs of large-scale batch production.

[0007] The technical solution of the present invention is:

[0008] A calibration method for a flexible tactile sensor, comprising:

[0009] Acquiring response data of the flexible tactile sensor array, performing feature extraction and data preprocessing on the response data to obtain standardized training data;

[0010] Based on the standardized training data, the sensor spatial features are extracted by a convolutional neural network, and the temporal response features are obtained by processing the data through a long short-term memory network. The obtained sensor spatial features and temporal response features are used as input to train and generate a calibration model;

[0011] Collecting environmental parameters of the flexible tactile sensor, and adaptively adjusting the calibration model according to the environmental parameters using a Bayesian optimization algorithm to obtain optimized calibration parameters;

[0012] The flexible tactile sensor is transported to a multi-station parallel calibration unit, and the flexible tactile sensor is parallel calibrated using the optimized calibration parameters to generate calibration data to obtain a calibrated flexible tactile sensor;

[0013] The calibrated flexible tactile sensor is subjected to a performance test, and production parameters and test results are recorded based on the calibration data to generate a unique identification code for quality traceability.

[0014] Preferably, the step of performing feature extraction and data preprocessing on the response data to obtain standardized training data includes:

[0015] Based on the response data, extracting signal features using wavelet transform to obtain a feature vector;

[0016] Performing principal component analysis and dimensionality reduction processing on the feature vector to eliminate data redundancy and obtain feature data after dimensionality reduction;

[0017] The feature data after dimensionality reduction is standardized to obtain the standardized training data.

[0018] Preferably, the extracting of sensor spatial features by a convolutional neural network and obtaining temporal response features by a long short-term memory network processing include:

[0019] Processing the standardized training data using multiple convolutional layers and pooling layers to obtain spatial feature data;

[0020] Inputting the spatial feature data into a long short-term memory network unit, extracting temporal features, and obtaining temporal feature data;

[0021] The spatial feature data and the temporal feature data are subjected to feature fusion to obtain the spatiotemporal features of the sensor.

[0022] Preferably, the training to generate a calibration model includes:

[0023] Dividing the standardized training data into a training set and a validation set according to a preset ratio;

[0024] Iteratively training the training set using an optimization algorithm until a preset training accuracy threshold is reached to obtain an initial calibration model;

[0025] The performance of the initial calibration model is evaluated using the validation set, and when the evaluation index meets the preset conditions, the calibration model is obtained.

[0026] Preferably, the collecting of environmental parameters of the flexible tactile sensor includes:

[0027] Multiple sensor nodes are set up in the calibration area to collect environmental data in real time using wireless transmission;

[0028] Configuring multiple types of sensors for each sensor node, wherein the multiple types of sensors collect temperature, humidity and air pressure data to obtain raw environmental data;

[0029] The original environmental data is filtered to eliminate noise interference and obtain the environmental parameters.

[0030] Preferably, the adaptive adjustment of the calibration model according to the environmental parameters using a Bayesian optimization algorithm includes:

[0031] Based on the environmental parameters, generating an optimization space of calibration parameters;

[0032] Using a Bayesian optimization algorithm to search for an optimal parameter combination in the optimization space to obtain optimized parameters;

[0033] The calibration model is adjusted according to the optimization parameters, and the optimized calibration parameters are obtained through filtering.

[0034] Preferably, the step of transporting the flexible tactile sensor to a multi-station parallel calibration unit includes:

[0035] Using a visual positioning system to identify the position of the flexible tactile sensor to obtain position information;

[0036] Based on the position information, the flexible tactile sensor is transferred to a calibration station by a manipulator; wherein a closed-loop control strategy is adopted during the transfer process to adjust the position of the flexible tactile sensor by monitoring the position of the sensor in real time and comparing it with a preset calibration position so that its position accuracy meets a preset range;

[0037] The flexible tactile sensor is fixed on a calibration station using a fixture.

[0038] Preferably, the parallel calibration of the flexible tactile sensor using the optimized calibration parameters includes:

[0039] Setting pressure, temperature, and humidity excitation conditions applied to the flexible tactile sensor as external stimuli according to the optimized calibration parameters to simulate its working conditions in different actual environments;

[0040] Synchronously collecting response signals of the plurality of flexible tactile sensors using a data acquisition module to obtain collected response signals;

[0041] The collected response signal is processed and feature extracted to obtain the calibration data.

[0042] Preferably, the performance test of the calibrated flexible tactile sensor includes:

[0043] Based on preset performance evaluation indicators, the performance parameters of the calibrated flexible tactile sensor are measured to obtain parameter test results;

[0044] Testing the response characteristics of the calibrated flexible tactile sensor under a variety of different environmental conditions to obtain response test results;

[0045] Based on the parameter test results and the response test results, and in accordance with a preset performance grading standard, the performance grade of the calibrated flexible tactile sensor is determined.

[0046] The present invention also provides a calibration device for a flexible tactile sensor, comprising:

[0047] A data acquisition module is used to obtain response data of the flexible tactile sensor array, perform feature extraction and data preprocessing on the response data, and obtain standardized training data;

[0048] a calibration model generation module, configured to extract sensor spatial features using a convolutional neural network based on the standardized training data, obtain temporal response features through processing using a long short-term memory network, and train and generate a calibration model using the obtained sensor spatial features and temporal response features as input;

[0049] An adaptive optimization module, configured to collect environmental parameters of the flexible tactile sensor, and adaptively adjust the calibration model according to the environmental parameters using a Bayesian optimization algorithm to obtain optimized calibration parameters;

[0050] a calibration execution module, configured to transport the flexible tactile sensor to a multi-station parallel calibration unit, perform parallel calibration on the flexible tactile sensor using the optimized calibration parameters, generate calibration data, and obtain a calibrated flexible tactile sensor;

[0051] The quality management module is used to perform performance testing on the calibrated flexible tactile sensor, record production parameters and test results based on the calibration data, and generate a unique identification code for quality traceability.

[0052] The beneficial effects of the present invention are:

[0053] 1. This invention adopts a deep learning model that combines convolutional neural networks and long short-term memory networks to build an intelligent calibration system, which realizes the accurate mapping between sensor signals and physical quantities, shortens the traditional calibration time from several hours to within 60 seconds, and controls the calibration error within ±2%.

[0054] 2. The portable multifunctional integrated calibration device designed by the present invention combines MEMS technology and has a volume of only 150×150×50mm 3 , integrating a micro excitation source, high-precision data acquisition and wireless communication system, greatly improving the flexibility and efficiency of the calibration process.

[0055] 3. The adaptive environmental simulation and compensation technology of the present invention monitors environmental parameters in real time and dynamically adjusts calibration conditions. It combines the Bayesian optimization algorithm to automatically generate the optimal calibration path, enabling the sensor to maintain stable performance in the temperature range of -10°C to 60°C and the humidity range of 20%-95% RH, greatly improving the environmental adaptability of the sensor.

[0056] 4. The multi-station parallel calibration unit design of the present invention integrates 12-16 independent calibration stations, realizing high-throughput calibration of the sensor array. The calibration capacity per unit time reaches 960 pieces / hour, meeting the production capacity requirement of 3 million pieces / year, significantly improving production efficiency.

[0057] 5. The present invention adopts a quality traceability system based on blockchain technology, assigns a unique ID to each sensor, records production parameters, calibration data and test results, realizes accurate traceability and analysis of quality problems, and improves product yield. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0059] Figure 1 This is a flow chart of a calibration method for a flexible tactile sensor of the present invention;

[0060] Figure 2 This is a schematic diagram of the calibration model architecture based on deep learning in the present invention;

[0061] Figure 3 The figure is a schematic structural diagram of a calibration device for a flexible tactile sensor according to the present invention. DETAILED DESCRIPTION

[0062] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0063] like Figure 1 As shown, the present invention provides a calibration method for a flexible tactile sensor, comprising the following steps:

[0064] Step S1, obtaining response data of the flexible tactile sensor array, performing feature extraction and data preprocessing on the response data to obtain standardized training data;

[0065] The flexible tactile sensor array was tested under various environmental conditions, collecting extensive response data. These included temperatures ranging from 0-60°C, humidity ranging from 20% to 90% RH, and various operational conditions, such as static pressure, dynamic pressure, and shear force. For each sensor unit, 100-1000 sets of data points were collected, including input stimulus values, output signal values, and corresponding environmental parameters. To ensure data quality and representativeness, Latin Hypercube sampling was used to ensure uniform data distribution across various environmental conditions and operational conditions.

[0066] After acquiring the response data, effective feature extraction and data preprocessing are required. First, the original signal is subjected to multi-scale analysis using wavelet transform to extract signal features containing the sensor response characteristics and form feature vectors. These features can effectively characterize performance parameters such as sensor sensitivity, linearity, hysteresis, drift, and cross-sensitivity. Next, the extracted feature vectors are subjected to principal component analysis (PCA) dimensionality reduction processing to eliminate redundant information in the data, reduce the subsequent computational effort, and retain the main feature information of the data. After dimensionality reduction, the feature data is normalized using the Z-score standardization method to make data of different dimensions comparable. Outlier detection methods are then used to identify and process anomalous data points to further improve data quality. Ultimately, standardized training data is obtained, providing a high-quality data foundation for the subsequent training of deep learning models.

[0067] like Figure 2 As shown, in step S2, based on the standardized training data, the sensor spatial features are extracted by a convolutional neural network, and the temporal response features are obtained by processing them through a long short-term memory network. The obtained sensor spatial features and temporal response features are used as input to train and generate a calibration model;

[0068] A hybrid deep learning architecture was designed that combines the strengths of convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) to simultaneously process both spatial and temporal sensor features. First, standardized training data is fed into a CNN structure consisting of three to five convolutional and pooling layers. The convolutional layers convolve the input data with kernels to extract local spatial features from the sensor array. The pooling layers downsample the feature maps to reduce the dimensionality of the feature maps, improving computational efficiency and enhancing the model's anti-interference capabilities. After multiple layers of convolution and pooling, feature maps representing the spatial distribution characteristics of the sensors are generated, referred to as spatial feature data. Next, the extracted spatial feature data is fed into an LSTM network. The LSTM network consists of two to three layers of LSTM units, each containing three control units: an input gate, a forget gate, and an output gate. This effectively captures the temporal dependencies of sensor responses. Using a memory mechanism, the LSTM network learns the sensor response characteristics at different time points, analyzes the temporal evolution of sensor signals, captures the dynamic response characteristics of the sensors, and ultimately outputs temporal feature data.

[0069] In the calibration system of the present invention, the specific structural design of the convolutional neural network (CNN) fully considers the extraction requirements of the spatial features of the sensor array. The CNN structure includes a deep network consisting of 3 main convolutional layers and 2 pooling layers. The first convolutional layer is configured with 32 convolution kernels of size 3×3, with a step size of 1 and a padding of 1 to maintain the size of the feature map, and uses the ReLU activation function to introduce nonlinear characteristics; the subsequent maximum pooling layer uses a 2×2 pooling window and a step size of 2 to halve the size of the feature map while retaining significant features. The second convolutional layer contains 64 3×3 convolution kernels, also using a parameter setting of step size 1, padding of 1 and the ReLU activation function to extract higher-level feature representations; the second pooling layer that follows adopts the same 2×2 maximum pooling strategy. The third convolutional layer is further increased to 128 3×3 convolution kernels to capture more complex spatial feature patterns, and a global average pooling layer is used to convert the feature map into a fixed-length feature vector. This progressively deepening convolution structure design can effectively extract spatial distribution features of different scales from sensor array data, providing a reliable basis for subsequent time series analysis.

[0070] The Long Short-Term Memory (LSTM) network is specifically optimized for the temporal response characteristics of sensors. This structure comprises two bidirectional LSTM layers, each with 128 hidden units, forming a deep recurrent network architecture. The LSTM unit uses a tanh activation function to process cell state updates, and a sigmoid function to control the on / off states of the three gating units (input gate, forget gate, and output gate), effectively modeling long-term dependencies. The network's bidirectional design enables it to simultaneously consider both historical and future state information about sensor responses, improving the accuracy of temporal pattern recognition. The temporal processing length is set to 64 time steps, a parameter verified through extensive experiments to effectively balance computational complexity and the ability to capture the dynamic response characteristics of sensors. During the feature fusion stage, an attention mechanism is used to weightedly fuse the spatial features extracted by the CNN and the temporal features extracted by the LSTM. This allows the model to adaptively focus on the importance of different features, further improving calibration accuracy.

[0071] Model training data is obtained through carefully designed data collection experiments, including comprehensive response data sets collected from different types and batches of flexible tactile sensors under various environmental conditions (temperature range 0-60°C, humidity range 20%-90% RH) in a strictly controlled experimental environment. The Latin hypercube sampling method is used to ensure uniform sampling within the environmental parameter space, with a total number of sampling points of no less than 10,000, covering the characteristics of the sensor under various working states and environmental conditions. Each sampling point records multi-dimensional information including the precise force value applied, temperature and humidity conditions, the original output signal of the sensor, and reference measurement values, forming a high-quality training data foundation. In the data preprocessing stage, the raw data is subjected to outlier detection, missing value processing, and noise filtering, and then the data is normalized using the Z-score normalization method to improve training efficiency and model generalization ability.

[0072] The model training process utilizes finely tuned hyperparameter settings to achieve optimal training results. The batch size is set to 64, achieving a good balance between training efficiency and gradient estimation accuracy. The initial learning rate is 0.001, and a cosine annealing strategy is used for dynamic adjustment, maintaining a high learning rate for rapid convergence in the early stages of training, and reducing the learning rate for fine-tuning in the later stages. To prevent overfitting, multiple regularization strategies are introduced, including an L2 regularization coefficient set to 0.0001 and a dropout layer with a ratio of 0.5. The total number of training iterations is 5,000, and an early stopping mechanism is used to automatically terminate training if performance on the validation set no longer improves. The entire training process is completed on a high-performance computing cluster equipped with Tesla V100 GPUs, with a typical training time of approximately 8 hours. After training, the model is thoroughly evaluated on a test dataset to ensure that the calibration performance meets the ±2% accuracy requirement under various conditions.

[0073] To fully utilize the spatiotemporal characteristics of sensors, this paper employs feature fusion technology to fuse the spatial feature data extracted by CNN with the temporal feature data extracted by LSTM to generate comprehensive spatiotemporal features of the sensors. Fusion methods include feature concatenation and weighted fusion. Feature concatenation directly concatenates two types of features to form a higher-dimensional feature vector, while weighted fusion assigns different weights to the two types of features based on their importance, and then generates a fused feature through weighted summation.

[0074] When training the calibration model, the standardized training data is first split into a training set and a validation set at an 80%:20% ratio. The hybrid deep learning model is trained using the training set, and the Adam optimizer is used to update the model parameters. The mean squared error (MSE) loss function measures the difference between the model's predicted values ​​and the true values. During training, dynamic learning rate adjustment, batch normalization, and dropout techniques are used to prevent overfitting and improve the model's generalization. Training is repeated for at least 5,000 times until the loss function value drops below a preset threshold, resulting in an initial calibration model.

[0075] The validation set is then used to evaluate the performance of the initial calibration model, using metrics including prediction mean square error (MSE), relative standard deviation (RSD), and computation time. The model is considered to meet performance requirements when the MSE is less than 0.5%, the RSD is less than 1%, and the computation time for a single-chip sensor is less than 50ms. If performance does not meet the requirements, the model is optimized by adjusting the network structure, optimizing hyperparameters (such as learning rate, batch size, number of hidden layer neurons, etc.), and introducing regularization methods (such as L1 and L2 regularization) until performance requirements are met, ultimately obtaining a calibrated model. To facilitate model deployment in embedded systems, model pruning and quantization techniques are also used to reduce the model size, thereby reducing the computational complexity and storage requirements of the model while ensuring model performance.

[0076] Finally, the trained deep learning model was combined with traditional temperature compensation algorithms and environmental correction methods to construct a comprehensive calibration model. A reinforcement learning mechanism was introduced, enabling the model to automatically adjust calibration parameters based on changes in environmental conditions, enabling adaptive calibration across a wide range of environmental conditions (e.g., -10°C to 60°C temperature range and 20%-95% RH humidity range). This comprehensive calibration model can control the sensor's measurement error within ±2% under various environmental conditions, significantly exceeding the ±5% accuracy achieved by traditional calibration methods.

[0077] Step S3, collecting environmental parameters of the flexible tactile sensor, and using a Bayesian optimization algorithm to adaptively adjust the calibration model according to the environmental parameters to obtain optimized calibration parameters;

[0078] It is necessary to design a complete environmental monitoring system to collect the environmental parameters of the flexible tactile sensor in real time. The system sets up multiple sensor nodes in the calibration area to form a distributed environmental monitoring network. Each sensor node is equipped with a high-precision temperature sensor (accuracy ±0.1°C), a humidity sensor (accuracy ±1%RH), an air pressure sensor (accuracy ±10Pa) and an electromagnetic field detector (frequency range 10Hz-1GHz). These sensors collect environmental data at a sampling frequency of 10Hz and transmit the data to the central controller in real time via wireless transmission. The wireless transmission uses low-power wide area network technology (such as LoRa or NB-IoT) to ensure transmission stability and low latency.

[0079] After acquiring the raw environmental data, preprocessing is required to eliminate noise. First, a Kalman filter algorithm is used to filter the temperature, humidity, and pressure data to eliminate random fluctuations and measurement noise. The data is then further purified using methods such as median filtering and wavelet denoising to ensure the accuracy of environmental parameters. Furthermore, the system monitors the changing trends of environmental parameters and predicts short-term changes in environmental conditions, providing forward-looking guidance for the calibration process.

[0080] Based on the acquired environmental parameters, the present invention uses a Bayesian optimization algorithm to adaptively adjust the calibration model. First, based on the current environmental parameters, an optimization space for the calibration parameters is generated. This optimization space contains multiple dimensions, such as temperature compensation coefficient, humidity correction factor, pressure excitation range, loading rate and other parameters. Then, the Bayesian optimization algorithm searches for the optimal parameter combination in this multidimensional parameter space. Compared with traditional grid search and random search, Bayesian optimization can more efficiently find the global optimal solution by constructing a probabilistic model of the parameter space (usually a Gaussian process), greatly reducing the number of trials in the search process.

[0081] The Bayesian optimization process consists of several steps: first, initializing the Gaussian process model based on historical calibration data; then, using an acquisition function (such as the expected improvement function (EI) or the upper confidence bound (UCB)) to determine the next parameter point to be evaluated; then, using this set of parameters to conduct actual calibration experiments and record the results; finally, updating the Gaussian process model, repeating the above process until a preset number of iterations is reached or convergence conditions are met. In this way, the algorithm can quickly find a near-optimal parameter combination within a limited number of experiments.

[0082] After obtaining the optimized parameters, they undergo further processing, including smoothing to prevent sudden changes, checking parameter ranges to ensure they are within a reasonable range, and verifying inter-parameter consistency to avoid conflicts. This process results in the final optimized calibration parameters, which are used to guide subsequent calibration processes. This allows the calibration model to automatically adjust parameters based on changing environmental conditions, achieving environmentally adaptive calibration and ensuring the sensor maintains high-precision measurement capabilities in a variety of environmental conditions.

[0083] Step S4, transporting the flexible tactile sensor to a multi-station parallel calibration unit, performing parallel calibration on the flexible tactile sensor using the optimized calibration parameters, generating calibration data, and obtaining a calibrated flexible tactile sensor;

[0084] An automated loading and conveying system transports the flexible tactile sensor to a multi-station parallel calibration unit. This process begins with a high-precision visual positioning system identifying the sensor's position. Equipped with a high-resolution industrial camera (resolution better than 5 microns), the vision system uses image processing algorithms to identify the sensor's edges, positioning marks, and electrode areas, accurately determining the sensor's spatial position and posture. Position recognition accuracy reaches ±0.05mm, ensuring precise grasping and placement.

[0085] Based on the acquired position information, a precision robot grasps the sensor from the conveyor belt and transfers it to the appropriate calibration station. The transfer process utilizes a closed-loop control strategy, using encoder feedback and a visual assistance system to monitor the sensor's position in real time, compare it with the preset calibration position, and dynamically adjust the robot's motion trajectory to ensure accurate sensor placement at the calibration station. To prevent mechanical damage to the flexible sensor during transfer, the robot grasps the sensor using controlled air pressure adsorption. The gripping force is intelligently adjusted based on the sensor's size and weight, maintaining a controllable range of 0.1-1.0N.

[0086] Once the sensor arrives at the calibration station, it is secured in place using a dedicated fixture. The fixture's design takes into account the characteristics of flexible sensors, employing edge clamping to avoid excessive pressure on sensitive sensor areas. Furthermore, the fixture features a fine-tuning mechanism that allows precise adjustment of the sensor's position within a ±0.1mm range, ensuring precise alignment between the sensor and the calibration device. The securing process, performed collaboratively by the robot and fixture, takes less than three seconds, significantly improving production efficiency.

[0087] In the multi-station parallel calibration unit, each station is equipped with a complete calibration system, including an excitation source module, a data acquisition module, and an environmental control module. According to the optimized calibration parameters obtained in step S3, the system automatically sets various working parameters. First, the excitation source module applies specific pressure, temperature, and humidity conditions to the sensor according to the parameter settings to simulate its working state in the actual use environment. Pressure excitation is applied by a pressure head driven by a precision piezoelectric actuator or a micro motor, with a range of 0.1N-50N and an accuracy of ±0.1N; temperature control is achieved through a micro Peltier element, with a range of 0-100℃ and an accuracy of ±0.2℃; humidity regulation is achieved through a micro ultrasonic atomizer and a solid desiccant switching system, with a range of 20%-95%RH and an accuracy of ±1.5%RH.

[0088] While the stimulus is applied, the data acquisition module simultaneously collects the sensor's response signals. The acquisition system utilizes a 24-bit, high-precision analog-to-digital converter with a sampling rate of 100kHz and a signal-to-noise ratio better than 100dB, accurately capturing even subtle changes in the sensor's response. The system supports simultaneous multi-channel acquisition, enabling simultaneous processing of multiple sensing units in the array. The collected data is transmitted to the central processing unit via a high-speed data bus for real-time processing and analysis.

[0089] The collected response signals are processed and feature extracted, including steps such as signal filtering, time domain feature extraction, frequency domain analysis, and statistical feature calculation. The processed data is compared with the preset standard values, the compensation coefficients and correction parameters are calculated, and a calibration data packet is generated. The entire calibration process is coordinated by a central control system, and multiple stations work in parallel, greatly improving calibration efficiency. For a sensor array with 64 sensing units, the calibration time for a single station is controlled within 60 seconds. In a system equipped with 16 calibration stations, the theoretical calibration capacity can reach 960 pieces / hour, meeting the needs of large-scale production.

[0090] After calibration, the sensor is transported to the next process, and the calibration data is saved in the central database and associated with the unique ID of the corresponding sensor, providing data support for subsequent quality traceability.

[0091] Step S5: performing a performance test on the calibrated flexible tactile sensor, recording production parameters and test results based on the calibration data, and generating a unique identification code for quality traceability.

[0092] After calibration, flexible tactile sensors undergo comprehensive performance testing to verify calibration validity and ensure product quality. Performance testing is divided into three tiers: basic electrical parameter testing, functional performance testing, and environmental adaptability testing. First, basic electrical parameter testing examines the sensor's fundamental electrical parameters, such as resistance, capacitance, and leakage current, to ensure that its electrical characteristics meet design specifications. This testing utilizes a high-precision LCR meter and ammeter, achieving a measurement accuracy of 0.1%, providing a foundational guarantee for sensor performance.

[0093] Next, the functional performance test focuses on evaluating the sensor's core performance indicators, including sensitivity, linearity, hysteresis, response time, and repeatability. This test uses standardized test equipment to apply a standard stimulus to the sensor under controlled conditions and record its response characteristics. The sensitivity test evaluates the sensor's responsiveness to external stimuli, requiring a sensitivity deviation of no more than ±5%. The linearity test evaluates the sensor's linear response characteristics within the measurement range, requiring a nonlinear error of less than 3%. The hysteresis test evaluates the sensor's hysteresis effect through load-unload cycles, requiring a hysteresis error of less than 2%. The response time test evaluates the sensor's response speed to rapidly changing stimuli, requiring a response time of less than 10ms. The repeatability test evaluates the sensor's stability through multiple measurements under the same conditions, requiring a repeatability error of less than 1%.

[0094] To verify the sensor's performance in real-world applications, the system also performs environmental adaptability testing. This test evaluates the sensor's response characteristics under a variety of environmental conditions (such as temperature, humidity, and air pressure), verifying the calibration's environmental adaptability. Test conditions cover a temperature range of -10°C to 60°C and a humidity range of 20% to 95% RH, ensuring the sensor maintains reliable performance in a variety of real-world environments. These tests provide a comprehensive assessment of the calibration's effectiveness and the sensor's performance.

[0095] Based on the test results, sensors are assigned performance grades. Based on pre-defined performance grading standards, sensors are divided into different levels, such as high-precision, standard, and economy, to meet the needs of different application scenarios. Performance grade information is recorded in the product database and associated with the physical product via a QR code or RFID tag for easy customer identification and use.

[0096] Quality traceability is crucial for ensuring product reliability and quickly locating issues. This system establishes a full-lifecycle quality traceability mechanism based on blockchain technology. The system assigns a universally unique identification code (UUID) to each sensor and records all sensor-related information on the blockchain, including raw material batches, production process parameters, calibration data, test results, and quality grades. This data is stored in encrypted form and cannot be tampered with, ensuring its authenticity and reliability. By scanning the QR code on the sensor or reading the embedded RFID tag, users or managers can access the sensor's complete history, enabling full-process traceability.

[0097] Furthermore, statistical analysis of this data is conducted to identify key factors affecting product quality and potential risk points, providing data support for production process optimization and quality improvement. Through continuous data collection and analysis, combined with statistical process control (SPC) methods, the system enables real-time monitoring of the production process and early warning, preventing batch quality issues. This closed-loop quality management mechanism has effectively improved product yield, gradually increasing it from an initial 85% to over 95%, exceeding the industry average of 90%.

[0098] The production management platform of the present invention adopts a modern software architecture design to build a fully functional flexible tactile sensor production management system. The platform is based on the B / S (browser / server) architecture, and the back-end is developed using the SpringBoot framework to provide high-performance REST API services; the front-end uses Vue.js to implement a responsive user interface, supporting cross-platform and multi-terminal access; the data storage layer adopts a hybrid architecture that combines MySQL relational database and MongoDB non-relational database, which are used to process structured production parameters and unstructured sensor characteristic curves, images and other data respectively. The system adopts a microservice architecture to decouple functional modules into independent services, and achieves efficient communication and load balancing through message queues (RabbitMQ) and service gateways (Spring CloudGateway), ensuring the stable operation of the system in a high-concurrency production environment. The platform is deployed in a containerized environment (Docker), supports elastic expansion and rapid deployment, and meets the needs of production lines of different sizes.

[0099] The production management platform includes four core functional modules: the equipment monitoring module collects and calibrates the working status, temperature and humidity parameters, and energy consumption data of the calibration equipment in real time, realizes millisecond-level status updates through WebSocket technology, and automatically issues warnings when equipment parameters deviate from the normal range; the production scheduling module realizes production planning, work order decomposition and resource allocation, adopts heuristic algorithms to optimize production scheduling, maximize equipment utilization, and display real-time production progress through electronic dashboards; the quality control module integrates the test data of automatic detection equipment, records the performance parameters and qualified status of each sensor, and generates detailed quality reports and statistical analysis; the data analysis module is the intelligent core of the system, and discovers patterns and anomalies hidden in production data through various data mining algorithms.

[0100] The data analysis module uses a variety of advanced algorithms, including random forests, gradient boosting trees (XGBoost) and deep neural networks, to conduct a comprehensive analysis of production process data. The algorithm can identify key factors that affect product quality, such as ambient temperature fluctuations during calibration, the rate of application of the excitation force value, and differences in material batches, and sort these factors by degree of influence through methods such as feature importance analysis, correlation heat map and principal component analysis. The system uses time series prediction methods to analyze the changing trends of process parameters over time, predict possible quality fluctuations, and recommend the optimal parameter adjustment strategy. For example, through analysis, it was found that when the calibration temperature is within the range of 28±0.5℃, the sensor consistency is the best; when the excitation force loading rate is maintained at 0.2N / s, the hysteresis is the smallest. These findings directly guide the optimization and adjustment of process parameters, making product performance more stable and consistent.

[0101] The yield improvement mechanism is a key component of the production management platform, employing a closed-loop quality control strategy to continuously improve production processes. This mechanism, based on the Plan-Do-Check-Act (PDCA) cycle, first uses Pareto analysis to identify key yield-limiting defects, such as uneven sensitivity, out-of-tolerance linearity, and zero drift. Fault Tree Analysis (FTA) and Failure Mode and Effects Analysis (FMEA) are then used to conduct in-depth root cause analysis, including inappropriate material mixes, insufficient environmental control accuracy, or inappropriate calibration parameter settings. Targeted improvement measures are then developed and validated. Finally, effective measures are standardized and incorporated into production specifications. The system also incorporates statistical process control (SPC) methods, providing real-time monitoring of key process parameters such as calibration temperature, force accuracy, and signal acquisition noise. Control charts are used to analyze parameter trends, automatically triggering alerts when parameters approach control limits or exhibit abnormal patterns. This comprehensive yield improvement mechanism has gradually increased the yield of sensor products from 85% at the project's initial stage to over 95%, significantly exceeding the industry average of 90%, while also improving product performance consistency and reliability.

[0102] Specifically, in step S1, the response data is subjected to feature extraction and data preprocessing to obtain standardized training data, including: based on the response data, using wavelet transform to extract signal features to obtain feature vectors; performing principal component analysis dimensionality reduction processing on the feature vectors to eliminate data redundancy and obtain feature data after dimensionality reduction; and performing standardization processing on the feature data after dimensionality reduction to obtain the standardized training data.

[0103] In a preferred embodiment, a large amount of sensor response data is collected from a flexible tactile sensor array under various environmental conditions (temperature 0-60°C, humidity 20%-90% RH) and different usage conditions (static pressure, dynamic pressure, shear force, etc.). Each sensor unit collects 100-1000 sets of data points, including input stimulus values, output signal values, and environmental parameters, to form an initial data set. Latin hypercube sampling is used during data collection to ensure uniform data distribution.

[0104] The collected sensor data is then analyzed for characteristics such as sensitivity, linearity, hysteresis, drift, and cross-sensitivity. Sensor signal features are extracted using methods such as wavelet transform and Fourier analysis. Dimensionality reduction techniques such as PCA and t-SNE are employed to reduce data redundancy. Z-score normalization and outlier detection are used to clean and preprocess the data to improve data quality.

[0105] In step S2, the spatial features of the sensor are extracted through a convolutional neural network and processed through a long short-term memory network to obtain the temporal response features, including: using multiple layers of convolutional layers and pooling layers to process the standardized training data to obtain spatial feature data; inputting the spatial feature data into a long short-term memory network unit to extract the temporal features to obtain temporal feature data; and performing feature fusion on the spatial feature data and the temporal feature data to obtain the spatiotemporal features of the sensor.

[0106] The training generates a calibration model, including: dividing the standardized training data into a training set and a validation set according to a preset ratio; iteratively training the training set using an optimization algorithm until a preset training accuracy threshold is reached to obtain an initial calibration model; using the validation set to perform performance evaluation on the initial calibration model, and when the evaluation index meets the preset conditions, the calibration model is obtained.

[0107] Specifically, during the training process of the calibration model, the standardized training data needs to be reasonably divided. The present invention adopts a stratified random sampling method to divide the data into a training set and a validation set according to a preset ratio (usually 80%:20%). Stratified sampling ensures that the distribution of various types of data in the training set and the validation set is consistent with the original data set, avoiding the model deviation that may be caused by the uneven distribution of samples. In actual operation, this ratio can be further adjusted according to the type of sensor and the amount of data. For example, for a large amount of data, a division ratio of 75%:25% can be adopted; for a small amount of data, a division ratio of 85%:15% can be adopted to ensure that the training set has enough samples to support the model to learn complex features.

[0108] Iterative training of the training set using an optimization algorithm is the core step of model construction. The present invention selects the Adam optimizer as the main optimization algorithm. This is because the Adam optimizer combines the advantages of the momentum method and RMSProp, can adaptively adjust the learning rate of each parameter, and is particularly suitable for processing non-stationary targets and data containing a lot of noise. Batch training is adopted in the training process, and the number of samples in each batch (batch size) is set to 64-128. This setting achieves a good balance between computational efficiency and optimization stability. To prevent overfitting, a variety of regularization techniques are also introduced in the training process, including L2 regularization (weight decay coefficient is set to 0.0001), dropout (dropout rate is 0.3-0.5) and early stopping mechanism. In addition, the learning rate adopts a dynamic adjustment strategy, and the initial value is set to 0.001. When the validation set loss no longer decreases for 5 consecutive epochs, the learning rate is reduced to 50% of the original value to accelerate model convergence. Iterative training continues until the mean absolute error (MAE) of the model on the training set drops below a preset threshold (usually 1%), or the maximum number of iterations (5000) is reached, at which point the initial calibration model is obtained.

[0109] After the initial calibration model training is completed, its performance needs to be comprehensively evaluated using the validation set. The evaluation adopts a multi-index comprehensive evaluation system, including multiple dimensions such as mean square error (MSE), mean absolute error (MAE), relative standard deviation (RSD), maximum error (Max Error) and calculation time. Among them, MSE and MAE evaluate the overall prediction accuracy of the model, RSD evaluates the stability of the prediction results, the maximum error evaluates the performance under extreme conditions, and the calculation time evaluates the practicality of the model. Specifically, the MSE is required to be less than 0.5%, MAE less than 0.8%, RSD less than 1%, the maximum error less than 2%, and the calculation time of a single-chip sensor less than 50ms. When all evaluation indicators meet the preset conditions at the same time, the model performance is considered qualified, and the final calibration model is obtained at this time.

[0110] If the evaluation results fail to meet the preset conditions, model optimization and adjustment are required. Adjustment strategies include: network structure adjustment (such as increasing or decreasing the number of network layers, adjusting the number of neurons), hyperparameter optimization (such as adjusting the learning rate, batch size, regularization coefficient), feature engineering improvement (such as increasing feature combinations, introducing new features), and integrated learning methods (such as model averaging, model fusion). The optimized and adjusted model is trained and evaluated again, and this cycle is repeated until a calibration model that meets the performance requirements is obtained. After the model is finalized, cross-validation (k-fold cross validation, k=5) will be used to further verify the robustness and generalization ability of the model to ensure that the model maintains stable performance under different data segmentation methods.

[0111] In order to enable the calibration model to be effectively deployed in the actual production environment, especially embedded in the embedded system of the portable calibration device, the present invention also optimizes the deployment of the final model. First, through the model pruning technology, the neuron connections that contribute less to the prediction results are removed to reduce the number of model parameters; secondly, through weight quantization, the 32-bit floating point number is converted to an 8-bit integer representation, which greatly reduces the model storage space requirement; finally, through model compilation and hardware acceleration optimization, the computing resources of the target platform are fully utilized to improve the model reasoning speed. After these optimizations, the model size is reduced by about 70%, the reasoning speed is increased by about 3 times, and the accuracy loss is controlled within 0.2%, which fully meets the actual application requirements. The final calibration model is not only high-precision and high-efficiency, but also has good deployment adaptability, laying a solid foundation for subsequent adaptive calibration work.

[0112] In a preferred embodiment, a hybrid deep learning architecture is designed, comprising a convolutional neural network (CNN) and a long short-term memory (LSTM) network. The CNN extracts spatial features from the sensor array and consists of three to five convolutional and pooling layers. The LSTM captures the sensor's temporal response characteristics and consists of two to three layers of LSTM units. The model inputs are raw sensor signals and environmental parameters, and the outputs are calibrated pressure values ​​and temperature compensation coefficients. The model is trained on a GPU cluster using the Adam optimizer and a mean squared error loss function, with at least 5,000 iterations.

[0113] Model performance was then evaluated using a cross-validation method, with the validation dataset containing 20% ​​sensor samples that were not used in training. Evaluation metrics included prediction mean squared error (MSE < 0.5%), relative standard deviation (RSD < 1%), and computation time (< 50ms / slice). Based on the validation results, the model's generalization capability was improved by adjusting the network structure, optimizing hyperparameters, and introducing regularization methods. Model pruning and quantization techniques were also used to reduce the model size, making it suitable for deployment in embedded systems.

[0114] A comprehensive calibration model was constructed by combining a trained deep learning model with traditional temperature compensation algorithms and environmental correction methods. The introduction of a reinforcement learning mechanism enabled the model to automatically adjust calibration parameters based on environmental changes, achieving adaptive calibration across a temperature range of -10°C to 60°C and a humidity range of 20% to 95% RH. The resulting model was able to control the sensor's measurement error to within ±2% under various environmental conditions, significantly exceeding the ±5% accuracy achieved by traditional calibration methods.

[0115] In step S3, the environmental parameters of the flexible tactile sensor are collected, including: setting up multiple sensor nodes in the calibration area and using wireless transmission to collect environmental data in real time; configuring multiple types of sensors for each sensor node, and the multiple types of sensors collect temperature, humidity and air pressure data to obtain raw environmental data; filtering the raw environmental data to eliminate noise interference and obtain the environmental parameters.

[0116] The method of adaptively adjusting the calibration model according to the environmental parameters using the Bayesian optimization algorithm includes: generating an optimization space for calibration parameters based on the environmental parameters; searching for an optimal parameter combination in the optimization space using the Bayesian optimization algorithm to obtain optimized parameters; adjusting the parameters of the calibration model according to the optimized parameters, and obtaining the optimized calibration parameters through filtering.

[0117] In a preferred embodiment, an integrated adaptive optimization module is designed, including a high-precision temperature sensor (±0.1°C), a humidity sensor (±1% RH), an air pressure sensor (±10Pa), and an electromagnetic field detector (frequency range 10Hz-1GHz). A distributed layout scheme is adopted, with multiple sensor nodes placed within the calibration area to form a three-dimensional distributed monitoring network for environmental parameters. Monitoring data is transmitted wirelessly to a central controller in real time with a sampling frequency of 10Hz, providing timely feedback for environmental parameter control.

[0118] Based on MEMS technology, an integrated micro-excitation source is designed. Pressure excitation uses a multi-stage pressure application mechanism composed of a piezoelectric driver and a micro-motor, with a force range of 0.1N-50N and a resolution of 0.05N. Temperature excitation uses a micro-Peltier element and a thin film resistor heater, with a temperature control range of 0-100°C and a heating / cooling rate of up to 5°C / s. Humidity control uses a micro-ultrasonic atomizer and a solid desiccant switching system, with a humidity adjustment range of 20%-95% RH. The entire excitation source module is controlled in a volume of 150×150×50mm 3 within the range.

[0119] Develop a precise environmental control system. Temperature control utilizes a thermoelectric cooler (TEC) combined with microfluidic cooling channels to achieve precise temperature control within a range of -10°C to 60°C within ±0.2°C. Humidity control combines ultrasonic humidification and molecular sieve adsorption drying technology to achieve precise humidity regulation within a range of 20% to 95% RH within ±1.5%. Air pressure control utilizes a micro air pump and precision pressure reducing valve to simulate an air pressure environment of 70kPa to 110kPa. The control system utilizes a PID algorithm combined with fuzzy logic, achieving a response time of less than 10 seconds and a stabilization time of less than 30 seconds.

[0120] An intelligent calibration strategy generation system has been developed to automatically generate the optimal calibration path and parameter settings based on real-time monitored environmental parameters and sensor characteristics. Using a Bayesian optimization algorithm, the system selects the most appropriate parameter combination from a preset calibration parameter space, including excitation force, loading rate, and temperature gradient. The system dynamically adjusts subsequent calibration steps based on initial test results, reducing redundant test points and optimizing the standard 50-point calibration process to 15-20 key points, significantly improving calibration efficiency while maintaining accuracy.

[0121] Using Bayesian networks and data fusion techniques, calibration data acquired under different environmental conditions is comprehensively analyzed. A model of the environmental dependence of sensor response is constructed, quantifying the influence of environmental factors such as temperature and humidity on sensor output. A multi-source data fusion algorithm generates a sensor response surface covering the entire environmental range, ensuring that calibration results remain valid under a wider range of environmental conditions. This ultimately results in a calibration data package containing compensation coefficients and correction formulas, ensuring that sensor measurement accuracy remains within ±3% over the temperature range of -10°C to 60°C and the humidity range of 20% to 95% RH.

[0122] In step S4, the flexible tactile sensor is transported to a multi-station parallel calibration unit, including: using a visual positioning system to identify the position of the flexible tactile sensor to obtain position information; based on the position information, the flexible tactile sensor is transported to the calibration station by a manipulator; wherein, a closed-loop control strategy is adopted during the transmission process, and the position of the flexible tactile sensor is adjusted by real-time monitoring of the position of the sensor and comparing it with a preset calibration position so that its position accuracy meets a preset range; and the flexible tactile sensor is fixed to the calibration station by a clamp.

[0123] The parallel calibration of the flexible tactile sensor using the optimized calibration parameters includes: setting pressure, temperature, and humidity excitation conditions applied to the flexible tactile sensor as external stimuli according to the optimized calibration parameters to simulate its working state in different actual environments; using a data acquisition module to synchronously acquire response signals of multiple flexible tactile sensors to obtain acquired response signals; and processing and feature extraction of the acquired response signals to obtain the calibration data.

[0124] In a preferred embodiment, a high-precision sensor array automatic loading system is developed, which uses visual positioning and pneumatic gripping technology to accurately place the sensors prepared in batches on the calibration station tray. The conveying system uses a dust-free conveyor belt and a precision positioning mechanism with a positioning accuracy of ±0.1mm to prevent positional offset and mechanical damage during the transmission process. The system supports sensor arrays of various sizes (5×5mm 2 Up to 100×100mm 2) automatic identification and adaptation processing.

[0125] A multi-station parallel calibration unit was designed. Each calibration unit contains 12-16 independent calibration stations, each equipped with a complete excitation, measurement, and environmental control system. A robotic automatic loading and unloading mechanism enables precise docking of sensors and calibration equipment. The calibration process is centrally scheduled by a central control system for optimal resource allocation. The complete calibration time for a single sensor array (64 sensing units) is controlled within 60 seconds. When all 16 stations are working in parallel, the system can calibrate 960 sensors per hour.

[0126] In step S5, the performance test of the calibrated flexible tactile sensor includes: measuring the performance parameters of the calibrated flexible tactile sensor based on preset performance evaluation indicators to obtain parameter test results; testing the response characteristics of the calibrated flexible tactile sensor under a variety of different environmental conditions to obtain response test results; and judging the performance level of the calibrated flexible tactile sensor based on the parameter test results and the response test results according to preset performance grading standards.

[0127] In a preferred embodiment, a multi-level quality inspection mechanism is introduced, including three levels: optical appearance inspection, electrical parameter testing, and functional performance testing. High-resolution industrial cameras (5-micron resolution) and automated testing equipment are used to detect physical defects and electrical performance of sensors, and qualification standards are set for key indicators such as sensitivity consistency, linearity, and hysteresis. A data traceability system based on blockchain technology is established, assigning a unique ID to each sensor, recording raw material information, production parameters, calibration data, and test results, and achieving full lifecycle quality traceability.

[0128] Develop a production management platform based on Industrial Internet of Things (IIoT) technology to integrate equipment status monitoring, production scheduling, materials management, and quality control. Introducing big data analytics technology, it mines and analyzes production process data, identifies key factors affecting product quality and production capacity, and provides decision support for process optimization. The system offers a web interface and mobile applications, supporting remote monitoring and management, enabling managers to understand and respond to production status in real time.

[0129] Based on big data analysis, a closed-loop mechanism for improving yield was established. By analyzing the distribution of defective product causes and identifying critical quality control points, production process parameters and calibration procedures were optimized in a targeted manner. Statistical Process Control (SPC) methods were introduced to monitor fluctuations in key process parameters, enabling early warning and intervention. Through continuous improvement and optimization, the yield rate of sensor products has been gradually increased from an initial 85% to over 95%.

[0130] like Figure 3As shown, the present invention also provides a calibration device for a flexible tactile sensor, comprising:

[0131] The data acquisition module 101 is used to obtain response data of the flexible tactile sensor array, perform feature extraction and data preprocessing on the response data, and obtain standardized training data;

[0132] A calibration model generation module 102 is configured to extract sensor spatial features based on the standardized training data through a convolutional neural network, obtain temporal response features through long short-term memory network processing, and use the obtained sensor spatial features and temporal response features as input to train and generate a calibration model;

[0133] An adaptive optimization module 103 is configured to collect environmental parameters of the flexible tactile sensor, and adaptively adjust the calibration model according to the environmental parameters using a Bayesian optimization algorithm to obtain optimized calibration parameters.

[0134] a calibration execution module 104, configured to transport the flexible tactile sensor to a multi-station parallel calibration unit, perform parallel calibration on the flexible tactile sensor using the optimized calibration parameters, generate calibration data, and obtain a calibrated flexible tactile sensor;

[0135] The quality management module 105 is used to perform performance testing on the calibrated flexible tactile sensor, record production parameters and test results based on the calibration data, and generate a unique identification code for quality traceability.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A calibration method for a flexible tactile sensor, characterized in that: include: Acquiring response data of the flexible tactile sensor array, performing feature extraction and data preprocessing on the response data to obtain standardized training data; Based on the standardized training data, the sensor spatial features are extracted by a convolutional neural network, and the temporal response features are obtained by processing the data through a long short-term memory network. The obtained sensor spatial features and temporal response features are used as input to train and generate a calibration model; Collecting environmental parameters of the flexible tactile sensor, and adaptively adjusting the calibration model according to the environmental parameters using a Bayesian optimization algorithm to obtain optimized calibration parameters; The flexible tactile sensor is transported to a multi-station parallel calibration unit, and the flexible tactile sensor is parallel calibrated using the optimized calibration parameters to generate calibration data to obtain a calibrated flexible tactile sensor; The calibrated flexible tactile sensor is subjected to a performance test, and production parameters and test results are recorded based on the calibration data to generate a unique identification code for quality traceability.

2. The method according to claim 1, characterized in that The feature extraction and data preprocessing of the response data to obtain standardized training data includes: Based on the response data, extracting signal features using wavelet transform to obtain a feature vector; Performing principal component analysis and dimensionality reduction processing on the feature vector to eliminate data redundancy and obtain feature data after dimensionality reduction; The feature data after dimensionality reduction is standardized to obtain the standardized training data.

3. The method according to claim 1, characterized in that The method of extracting sensor spatial features through a convolutional neural network and processing them through a long short-term memory network to obtain temporal response features includes: Processing the standardized training data using multiple convolutional layers and pooling layers to obtain spatial feature data; Inputting the spatial feature data into a long short-term memory network unit, extracting temporal features, and obtaining temporal feature data; The spatial feature data and the temporal feature data are subjected to feature fusion to obtain the spatiotemporal features of the sensor.

4. The method according to claim 1, wherein The training generates a calibration model, including: Dividing the standardized training data into a training set and a validation set according to a preset ratio; Iteratively training the training set using an optimization algorithm until a preset training accuracy threshold is reached to obtain an initial calibration model; The performance of the initial calibration model is evaluated using the validation set, and when the evaluation index meets the preset conditions, the calibration model is obtained.

5. The method according to claim 1, characterized in that The collecting of environmental parameters of the flexible tactile sensor includes: Multiple sensor nodes are set up in the calibration area to collect environmental data in real time using wireless transmission; Configuring multiple types of sensors for each sensor node, wherein the multiple types of sensors collect temperature, humidity and air pressure data to obtain raw environmental data; The original environmental data is filtered to eliminate noise interference and obtain the environmental parameters.

6. The method according to claim 1, characterized in that The adaptive adjustment of the calibration model according to the environmental parameters using the Bayesian optimization algorithm includes: Based on the environmental parameters, generating an optimization space of calibration parameters; Using a Bayesian optimization algorithm to search for an optimal parameter combination in the optimization space to obtain optimized parameters; The calibration model is adjusted according to the optimization parameters, and the optimized calibration parameters are obtained through filtering.

7. The method according to claim 1, characterized in that The method of transporting the flexible tactile sensor to a multi-station parallel calibration unit includes: Using a visual positioning system to identify the position of the flexible tactile sensor to obtain position information; Based on the position information, the flexible tactile sensor is transferred to a calibration station by a manipulator; wherein a closed-loop control strategy is adopted during the transfer process to adjust the position of the flexible tactile sensor by monitoring the position of the sensor in real time and comparing it with a preset calibration position so that its position accuracy meets a preset range; The flexible tactile sensor is fixed on a calibration station using a fixture.

8. The method according to claim 1, characterized in that The parallel calibration of the flexible tactile sensor using the optimized calibration parameters includes: Setting pressure, temperature, and humidity excitation conditions applied to the flexible tactile sensor as external stimuli according to the optimized calibration parameters to simulate its working conditions in different actual environments; Synchronously collecting response signals of the plurality of flexible tactile sensors using a data acquisition module to obtain collected response signals; The collected response signal is processed and feature extracted to obtain the calibration data.

9. The method according to claim 1, characterized in that The performance test of the calibrated flexible tactile sensor includes: Based on preset performance evaluation indicators, the performance parameters of the calibrated flexible tactile sensor are measured to obtain parameter test results; Testing the response characteristics of the calibrated flexible tactile sensor under a variety of different environmental conditions to obtain response test results; Based on the parameter test results and the response test results, and in accordance with a preset performance grading standard, the performance grade of the calibrated flexible tactile sensor is determined.

10. A calibration device for a flexible tactile sensor, characterized in that: include: A data acquisition module is used to obtain response data of the flexible tactile sensor array, perform feature extraction and data preprocessing on the response data, and obtain standardized training data; a calibration model generation module, configured to extract sensor spatial features using a convolutional neural network based on the standardized training data, obtain temporal response features through processing using a long short-term memory network, and train and generate a calibration model using the obtained sensor spatial features and temporal response features as input; An adaptive optimization module, configured to collect environmental parameters of the flexible tactile sensor, and adaptively adjust the calibration model according to the environmental parameters using a Bayesian optimization algorithm to obtain optimized calibration parameters; a calibration execution module, configured to transport the flexible tactile sensor to a multi-station parallel calibration unit, perform parallel calibration on the flexible tactile sensor using the optimized calibration parameters, generate calibration data, and obtain a calibrated flexible tactile sensor; The quality management module is used to perform performance testing on the calibrated flexible tactile sensor, record production parameters and test results based on the calibration data, and generate a unique identification code for quality traceability.

Images