Industrial robot food processing system based on multi-mode perception and intelligent regulation and control

Abstract

The invention belongs to the field of industrial automation, and provides a system based on multi-mode perception and intelligent regulation and control. By integrating various sensors on the robot and applying technologies such as multi-modal sensing fusion, bionics, real-time feedback, multi-source data fusion, reinforcement learning and block chains, accurate grabbing and sorting of food materials, dynamic regulation and control of parameters in the processing process, online quality monitoring, multi-robot collaborative scheduling and quality tracing are achieved. The invention aims to solve the problems of difficult material grabbing, inflexible parameter regulation and control, inaccurate quality monitoring and the like in food processing, improve the automation and intelligence level of the food processing industry, and meet the production requirements of high efficiency, safety and high quality.

Description

Technical Field

[0001] The present invention relates to the field of industrial robot automation, specifically focusing on the key technological innovations of industrial robots in the food processing process. Background Art

[0002] As an important industry to ensure people's daily life needs, the food processing industry has put forward higher requirements for the quality, safety and diversity of food with the improvement of people's living standards and the change of consumption concepts. Industrial robots, with their characteristics of automation, standardization and high efficiency, are increasingly widely used in food processing, covering multiple links such as food handling, sorting, cutting, and packaging. However, there are still many technical bottlenecks in the application of industrial robots in the food processing field, which limit the further development of the food processing industry.

[0003] Problems Existing in the Prior Art

[0004] Difficulties in Precise Grasping and Sorting of Food Materials: Food materials vary in shape, size, texture, and have complex surface characteristics, such as smooth surface, moisture, and easy deformation. Traditional industrial robot grasping and sorting systems mainly rely on simple visual recognition and fixed grasping strategies, which are difficult to accurately identify different types of food materials, and are prone to damage the materials during the grasping process, affecting the quality of food. At the same time, due to the randomness and diversity of food materials, existing sorting algorithms cannot be dynamically adjusted according to the real-time state and characteristics of the materials, resulting in low sorting efficiency and high error rates.

[0005] Difficulties in Dynamically Regulating Processing Parameters: In the food processing process, different food raw materials, processing technologies and environmental conditions have different requirements for processing parameters. Most existing industrial robot processing systems adopt fixed processing parameters and preset operation processes, and cannot be dynamically adjusted according to the real-time changes during the processing. For example, during the baking process, the temperature and time of the oven need to be precisely controlled according to the type, size and raw material composition of the food, but existing systems cannot perceive the changes of these factors in real time, resulting in unstable food quality, large differences in taste and appearance.

[0006] Low Reliability of On-line Monitoring of Product Quality: On-line monitoring of food quality is a key link to ensure food safety and quality. At present, industrial robots mainly rely on single detection means, such as visual detection or weight detection, in on-line monitoring of food quality. This detection method is easily affected by factors such as detection environment, food material characteristics and detection equipment accuracy, and cannot comprehensively and accurately detect food quality problems. Moreover, existing monitoring algorithms cannot deeply analyze and process detection data, and it is difficult to realize real-time evaluation and prediction of food quality, resulting in some unqualified products flowing into the market and there are potential food safety hazards. Summary of the Invention

[0007] The present invention provides an industrial robot food processing system based on multi-modal perception and intelligent regulation, including:

[0008] A precise grasping and sorting algorithm module for food materials based on multi-modal perception fusion and bionics. By integrating various sensors such as high-resolution vision sensors, tactile sensors, force sensors, and spectral sensors on an industrial robot, multi-modal information such as the image, shape, texture, hardness, and composition of food materials is obtained. The data is fused using multi-modal perception fusion technology, and an end effector for adaptive grasping is designed in combination with bionic principles to achieve precise grasping and sorting of food materials.

[0009] A dynamic regulation algorithm module for food processing process parameters based on real-time feedback and adaptive optimization. Temperature sensors, humidity sensors, pressure sensors, and composition sensors are installed on the processing equipment of the industrial robot to monitor the parameters and environmental factors during the processing process in real time. In combination with the adaptive optimization algorithm, dynamic regulation of the processing process parameters is achieved according to the real-time feedback data and the preset quality target.

[0010] An online food quality monitoring algorithm module based on multi-source data fusion and deep learning. Visual inspection equipment, infrared detection equipment, near-infrared detection equipment, and microbial detection equipment are installed on the food processing production line to obtain multi-source data such as the appearance, internal structure, composition, and microbial content of food. The data is fused using multi-source data fusion technology, and in combination with deep learning algorithms, such as the combination of convolutional neural network (CNN) and recurrent neural network (RNN), online monitoring and prediction of food quality are achieved.

[0011] A multi-robot collaborative food processing scheduling algorithm module based on reinforcement learning. Multiple industrial robots in the food processing workshop are regarded as a multi-agent system. The state space is defined to include the position of the robot, task progress, working status, and environmental information of the workshop, etc. The action space is defined to include various motion actions and task allocation decisions of the robot. A reward function is designed with processing efficiency, product quality, equipment utilization rate, etc. as reward indicators. Using reinforcement learning algorithms, such as deep Q-network (DQN) or proximal policy optimization algorithm (PPO), efficient collaborative processing scheduling among multiple robots is achieved.

[0012] A food processing quality traceability algorithm module based on blockchain technology. During the food processing process, raw material information, processing equipment parameters, processing process parameters, quality inspection data, etc. in the production process are collected in real time by sensors, encrypted and stored through blockchain technology, and in combination with smart contract technology, the whole process traceability of food quality is achieved.

[0013] Furthermore, in the precise grasping and sorting algorithm module of food materials based on multi-modal perception fusion and bionics, the multi-modal data fusion adopts a fusion network based on the attention mechanism.

[0014] Furthermore, in the dynamic regulation algorithm module of food processing process parameters based on real-time feedback and adaptive optimization, the established mathematical model of the food processing process considers the influence of food raw material characteristics, processing technology, and environmental factors on processing parameters.

[0015] Furthermore, in the online food quality monitoring algorithm module based on multi-source data fusion and deep learning, when performing fusion processing on multi-source data, normalization or standardization methods are used to preprocess different types of data.

[0016] Furthermore, in the multi-robot collaborative food processing scheduling algorithm module based on reinforcement learning, when assigning tasks, factors such as the load capacity of the robot, working range, and the urgency of the task are considered.

[0017] Furthermore, in the food processing quality traceability algorithm module based on blockchain technology, the smart contract defines the process of food quality traceability, including data query rules for stages such as raw material procurement, processing links, quality inspection, and product sales.

[0018] Furthermore, in the precise grasping and sorting algorithm module of food materials based on multi-modal perception fusion and bionics, when training the comprehensive feature model using deep learning algorithms, the cross-entropy loss function and Adam optimizer are used for parameter optimization.

[0019] Furthermore, in the dynamic regulation algorithm module of food processing process parameters based on real-time feedback and adaptive optimization, the adaptive optimization algorithm adopts an adaptive proportional-integral-derivative (PID) control algorithm or a model predictive control (MPC) algorithm.

[0020] Furthermore, in the online food quality monitoring algorithm module based on multi-source data fusion and deep learning, when training the deep learning model, the generalization ability of the model is improved by increasing the diversity of training data, adjusting the network structure and hyperparameters, etc.

[0021] Furthermore, in the multi-robot collaborative food processing scheduling algorithm module based on reinforcement learning, an experience replay mechanism is adopted to store and reuse samples of the interaction between the robot and the environment, improving the training efficiency and stability of the reinforcement learning algorithm.

[0022] Beneficial effects:

[0023] The key technologies of the present invention bring significant efficiency improvements to food processing. Based on algorithms of multi-modal perception fusion and bionics, integrating multi-sensor information and simulating human grasping actions, the grasping accuracy is increased from 70% to over 95%, the material damage rate is reduced to below 5%, the sorting efficiency is increased by about 40%, effectively guaranteeing food quality. The real-time feedback and adaptive optimization algorithm dynamically adjusts processing parameters according to real-time data, reducing the defective rate from 12% to below 3%, significantly improving the consistency of food taste and appearance, and enhancing market competitiveness. The multi-source data fusion and deep learning algorithm integrates data from various detection devices, increasing the detection accuracy from 75% to over 92%, predicting quality problems 2 - 3 hours in advance, and effectively reducing unqualified products. The multi-robot collaborative scheduling algorithm based on reinforcement learning increases the equipment utilization rate from 60% to over 85%, improves the processing efficiency by about 50%, and greatly reduces production costs. The quality traceability algorithm based on blockchain technology shortens the traceability time from several hours to several minutes, improves consumer trust, and enhances the market reputation of enterprises. It comprehensively promotes the food processing industry to move towards a new stage of high precision, high efficiency, and high reliability development. Description of the Drawings

[0024] Figure 1 Overall process schematic diagram. Detailed Implementation Manner

[0025] Example 1

[0026] Example of the precise grasping and sorting algorithm for food materials based on multi-modal perception fusion and bionics

[0027] Sensor deployment and data acquisition: On the end effector of the industrial robot, a high-resolution vision sensor with a resolution of 2000×1500 is installed vertically to obtain high-definition images of food materials; tactile sensors are evenly distributed on the grasping part of the end effector, capable of perceiving the texture and pressure changes on the material surface; force sensors are installed at the grasping joints with an accuracy of up to ±0.05N to monitor the grasping force in real time; spectral sensors are connected by optical fibers to quickly analyze the composition of food materials. Each sensor collects data in real time at a frequency of 100Hz and transmits it to the data processing unit through a high-speed USB 3.0 interface.

[0028] Multi-modal data fusion and feature model construction: The data processing unit uses a fusion network based on the attention mechanism to perform convolution operations on visual images to extract features such as shape and color; filter tactile data to remove noise interference; normalize force sensor data; extract features from spectral data to determine the material composition. The fusion network learns the importance of different modal data, assigns weights to each modal feature, and constructs a comprehensive feature model of food materials.

[0029] Bionic end effector design and grasping and sorting strategy implementation: According to the bionic principle, design an end effector with multiple movable joints and a soft grasping surface to simulate the flexible grasping action of human fingers. Use deep learning algorithms, such as the combination of convolutional neural network (CNN) and recurrent neural network (RNN), to learn and train the comprehensive feature model. In actual operation, input the multi-modal data collected in real time into the trained model, and the model automatically adjusts the grasping force, posture, and action of the end effector according to the material characteristics.

[0030] Modeling and solving process: In the model training stage, use the cross-entropy loss function to measure the difference between the prediction result and the true label, use the Adam optimizer, set the learning rate to 0.001, and perform iterative training 500 times. Continuously adjust the network parameters through the backpropagation algorithm to continuously improve the recognition accuracy of the food material characteristics and the accuracy of the grasping strategy of the model.

[0031] Efficiency improvement: The grasping accuracy of the traditional simple visual recognition and fixed grasping strategy is about 70%, and the material damage rate reaches 15%. The grasping accuracy of this algorithm is increased to more than 95%, the material damage rate is reduced to less than 5%, and the sorting efficiency is increased by about 40%, effectively ensuring the food quality.

[0032] Example of a dynamic regulation algorithm for food processing process parameters based on real-time feedback and adaptive optimization

[0033] Sensor installation and data acquisition: Install high-precision temperature sensors with an accuracy of ±0.5°C on industrial robot processing equipment, such as ovens and mixers, to monitor the processing temperature in real time; humidity sensors are used to monitor the environmental humidity; pressure sensors are installed on the processing components to measure the processing pressure; component sensors use near-infrared spectroscopy technology to analyze the food components in real time. Each sensor collects data in real time at a frequency of 20Hz and transmits it to the control system through industrial Ethernet.

[0034] Food processing process mathematical model establishment and adaptive optimization algorithm implementation: The control system establishes a mathematical model of the food processing process considering food raw material characteristics, processing technology, and environmental factors based on a large amount of experimental data and processing technology theory. Use the adaptive proportional-integral-differential (PID) control algorithm to calculate the adjustment amount of the processing parameters according to the real-time feedback data and the preset quality target. For example, during the baking process, when the temperature sensor detects that the temperature is lower than the set value, the PID controller calculates the heating power that needs to be increased according to the deviation.

[0035] Dynamic regulation and execution of processing parameters: The industrial robot adjusts processing parameters in real time according to the control instructions generated by the control system, such as processing time, temperature, pressure, speed, etc. During the processing, the processing parameters are continuously optimized dynamically based on newly collected data to ensure the stable quality of the processed food.

[0036] Modeling and solving process: When establishing a mathematical model, the parameters in the model are determined through fitting and analysis of a large amount of experimental data. The adaptive PID control algorithm adjusts the proportional coefficient (Kp), integral coefficient (Ki), and differential coefficient (Kd) according to the real-time deviation, enabling the processing parameters to quickly and accurately track the target value.

[0037] Efficiency improvement: The traditional fixed-parameter processing method leads to unstable food quality, with a defective rate of approximately 12%. This algorithm reduces the defective rate to less than 3%, significantly improving the consistency of the taste and appearance of the food and enhancing the market competitiveness.

[0038] Example of an online food quality monitoring algorithm based on multi-source data fusion and deep learning

[0039] Deployment of detection equipment and data collection: On the food processing production line, high-definition vision detection equipment is deployed in sequence to detect the appearance defects of food; infrared detection equipment is used to detect the internal structure of food; near-infrared detection equipment is used to analyze food components; and microbial detection equipment uses rapid detection technology to monitor the microbial content in real time. Each detection equipment collects data at a frequency of 30 Hz and transmits it to the data processing center through the network.

[0040] Multi-source data fusion and feature extraction: The data processing center uses a multi-source data fusion algorithm based on deep learning to fuse the data collected by different detection equipment. First, the data is preprocessed by normalization to eliminate the influence of data dimensions. Then, a convolutional neural network (CNN) is used to extract features from the fused data, extracting the key feature information of food quality.

[0041] Training of the deep learning model and execution of quality monitoring: Collect quality data of 5000 different types of food samples to construct a training dataset, covering normal food and various defective foods. A model combining a convolutional neural network (CNN) and a recurrent neural network (RNN) is used for training. The CNN is responsible for extracting static image and data features, and the RNN is used to process time series data, such as the change of microbial content over time. During the actual monitoring process, the extracted feature information is input into the trained model to achieve rapid and accurate evaluation and prediction of food quality.

[0042] Modeling and solution process: During model training, the mean squared error loss function is adopted, and the Stochastic Gradient Descent (SGD) algorithm is used. The learning rate is set to 0.01, the momentum is 0.9, and iterative training is carried out 800 times. By continuously adjusting the network parameters, the evaluation accuracy and prediction accuracy of the model for food quality are continuously improved.

[0043] Enhancement: The quality detection accuracy of traditional single detection means is about 75%, and potential quality problems cannot be detected in time. The detection accuracy of this algorithm is increased to more than 92%, and quality problems can be predicted 2 - 3 hours in advance, effectively reducing the production of unqualified products.

[0044] Embodiment of a multi - robot collaborative food processing scheduling algorithm based on reinforcement learning

[0045] Definition of state space, action space, and reward function: Clearly define the state space of the multi - robot collaborative food processing process, including the positions of the robots (represented by coordinates), task progress (the ratio of the number of completed tasks to the total number of tasks), working status (idle, busy, faulty, etc.), and environmental information of the workshop (temperature, humidity, etc.); define the action space, including the linear movement, rotation, and task assignment decision of the robots; design the reward function, taking processing efficiency (the number of tasks completed per unit time), product quality (the reciprocal of the defective rate), equipment utilization rate (the ratio of actual working time to total time), etc. as reward indicators. For example, +10 is rewarded for each successfully completed high - quality task, and -20 is punished for the appearance of defective products.

[0046] Reinforcement learning model training and application: Use the Proximal Policy Optimization algorithm (PPO) to learn and train the multi - robot collaborative food processing process. During training, the robots select actions according to the current state, interact with the environment, and obtain rewards and new state information. Through continuous iterative training, the robots gradually learn the optimal collaborative processing scheduling strategy. In actual production, the robots select actions according to the real - time state and adjust the strategy according to the reward feedback after executing the actions, realizing efficient collaborative work among multiple robots.

[0047] Modeling and solution process: The PPO algorithm samples the trajectories under the current policy, calculates the advantage function, and updates the policy network and value network using importance sampling. During training, the discount factor is set to 0.98, and the update step size is 30. Continuously optimize the policy network parameters to enable the robots to make optimal decisions according to different states.

[0048] Enhancement: The traditional static scheduling method results in an equipment utilization rate of only 60% and low processing efficiency. This algorithm increases the equipment utilization rate to more than 85% and improves the processing efficiency by about 50%, effectively reducing production costs.

[0049] Embodiment of Food Processing Quality Traceability Algorithm Based on Blockchain Technology

[0050] Data Collection and Storage: During the food processing process, sensors are used to collect raw material information (supplier, batch number, production date, etc.), processing equipment parameters (temperature, pressure, running time, etc.), processing process parameters (processing steps, time, temperature, etc.), and quality inspection data (appearance, composition, microbial content, etc.) in real time. These data are encrypted and stored through blockchain technology to form an immutable distributed ledger.

[0051] Smart Contract Establishment and Execution: According to the food processing process and quality traceability requirements, a smart contract is established to define the rules and processes for quality traceability. For example, when a consumer queries food information, the smart contract queries and returns the whole process information of the food from raw material procurement to final product sales based on the input product number. During the quality traceability process, through the distributed ledger and smart contract of the blockchain, rapid and accurate traceability of food quality is achieved.

[0052] Modeling and Solving Process: When establishing a smart contract, Solidity language is used to write code to define data structures and operation functions. Through the consensus mechanism of the blockchain, the consistency and security of the ledger data are ensured.

[0053] Efficiency Enhancement: Traditional quality traceability methods rely on manual records and paper documents, with long traceability times and prone to errors. This algorithm shortens the quality traceability time from several hours to several minutes, improves consumers' trust in food quality, and enhances the market reputation of enterprises.

Claims

1. An industrial robot food processing system based on multimodal perception and intelligent control, characterized in that: include: The algorithm module for accurate grasping and sorting of food materials based on multimodal perception fusion and bionics integrates multiple sensors such as high-resolution visual sensors, tactile sensors, force sensors, and spectral sensors on industrial robots to obtain multimodal information such as the image, shape, texture, hardness, and composition of food materials. The module uses multimodal perception fusion technology to fuse data and combines the principles of bionics to design an adaptive grasping end effector to achieve accurate grasping and sorting of food materials. The food processing parameters dynamic control algorithm module based on real-time feedback and adaptive optimization installs temperature sensors, humidity sensors, pressure sensors and ingredient sensors on the processing equipment of industrial robots to monitor the parameters and environmental factors in the processing process in real time. Combined with the adaptive optimization algorithm, the dynamic control of the processing parameters is achieved according to the real-time feedback data and the preset quality goals. The food quality online monitoring algorithm module based on multi-source data fusion and deep learning installs visual inspection equipment, infrared inspection equipment, near-infrared inspection equipment and microbial inspection equipment on the food processing production line to obtain multi-source data such as food appearance, internal structure, composition, microbial content, etc. The multi-source data fusion technology is used to fuse the data, combined with deep learning algorithms such as convolutional neural network (CNN) and recurrent neural network (RNN), to achieve online monitoring and prediction of food quality; The multi-robot collaborative food processing scheduling algorithm module based on reinforcement learning regards multiple industrial robots in the food processing workshop as a multi-agent system, defines the state space including the robot's position, task progress, working status and workshop environment information, defines the action space including the robot's various motion actions and task allocation decisions, designs the reward function and takes processing efficiency, product quality, equipment utilization rate and other indicators as reward indicators, and uses reinforcement learning algorithms such as deep Q network (DQN) or proximal policy optimization algorithm (PPO) to achieve efficient collaborative processing scheduling among multiple robots; The food processing quality traceability algorithm module based on blockchain technology uses sensors to collect real-time raw material information, processing equipment parameters, processing technology parameters, quality inspection data, etc. in the food processing process, and encrypts and stores them through blockchain technology. Combined with smart contract technology, it realizes the whole process traceability of food quality.

2. The industrial robot food processing system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the food material precise grasping and sorting algorithm module based on multimodal perception fusion and bionics, multimodal data fusion adopts a fusion network based on the attention mechanism.

3. The industrial robot food processing system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the food processing process parameter dynamic control algorithm module based on real-time feedback and adaptive optimization, the established food processing process mathematical model takes into account the influence of food raw material characteristics, processing technology and environmental factors on processing parameters.

4. The industrial robot food processing system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the food quality online monitoring algorithm module based on multi-source data fusion and deep learning, when fusing multi-source data, normalization or standardization methods are used to pre-process different types of data.

5. The industrial robot food processing system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the multi-robot collaborative food processing scheduling algorithm module based on reinforcement learning, factors such as the robot's load capacity, working range, and urgency of the task are considered when assigning tasks.

6. The industrial robot food processing system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the food processing quality traceability algorithm module based on blockchain technology, the smart contract defines the process of food quality traceability, including data query rules for stages such as raw material procurement, processing, quality inspection, and product sales.

7. The industrial robot food processing system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the food material precise grasping and sorting algorithm module based on multimodal perception fusion and bionics, when the comprehensive feature model is trained using a deep learning algorithm, a cross entropy loss function and an Adam optimizer are used for parameter optimization.

8. The industrial robot food processing system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the food processing process parameter dynamic control algorithm module based on real-time feedback and adaptive optimization, the adaptive optimization algorithm adopts an adaptive proportional-integral-differential (PID) control algorithm or a model predictive control (MPC) algorithm.

9. The industrial robot food processing system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the food quality online monitoring algorithm module based on multi-source data fusion and deep learning, when training the deep learning model, the generalization ability of the model is improved by increasing the diversity of training data, adjusting the network structure and hyperparameters, etc.

10. The industrial robot food processing system based on multimodal perception and intelligent control according to claim 1, characterized in that: In the multi-robot collaborative food processing scheduling algorithm module based on reinforcement learning, an experience replay mechanism is used to store and reuse samples of the robot's interaction with the environment, thereby improving the training efficiency and stability of the reinforcement learning algorithm.

Images