Industrial cleaning system based on multi-mode perception and intelligent regulation and control

Abstract

The invention belongs to the field of industrial automation, and discloses an industrial cleaning system based on multi-mode sensing and intelligent regulation and control. By integrating various sensors, accurate stain identification is realized by applying multi-modal sensing and spectral analysis; cleaning path intelligent planning is carried out by means of topological optimization and reinforcement learning; intelligent regulation and control of cleaning parameters are achieved through self-adaptive control and model prediction; a multi-robot collaborative cleaning system is constructed to process large-scale complex objects; and an intelligent monitoring and management platform is built. The invention aims to solve the problems of inaccurate stain identification, unreasonable path planning, unintelligent parameter regulation and control and the like in industrial cleaning, improve the efficiency, quality and adaptability of industrial cleaning and meet the requirements of modern industrial production.

Description

Technical Field

[0001] The present invention relates to the field of industrial robot automation, and particularly to an industrial cleaning system based on multi-modal perception and intelligent regulation. Background Art

[0002] Among the numerous links in industrial production, industrial cleaning is a key step to ensure product quality and extend the service life of equipment. Whether it is the industries of mechanical manufacturing, electronic processing, or food and beverage, there are strict requirements for industrial cleaning. With the development of industrial automation, industrial robots have gradually been applied to industrial cleaning tasks to replace traditional manual cleaning methods. However, there are still many problems with industrial robots in industrial cleaning at present, which seriously restrict their wide application and further development in this field.

[0003] There are a wide variety of objects to be cleaned in industrial production, and there are huge differences in the types, distributions, and adhesion degrees of surface stains. Traditional industrial robot cleaning systems mainly rely on simple visual recognition technology, which can only roughly judge the presence or absence of stains and cannot accurately identify the specific types, components, and distribution of stains. This makes it impossible for the robot to select appropriate cleaning processes and parameters according to the actual situation of the stains during the cleaning process, resulting in poor cleaning effects and even possible unnecessary damage to the cleaning objects.

[0004] Different cleaning objects have different shapes and structures, and different cleaning paths need to be adopted to ensure the comprehensiveness and uniformity of cleaning. Existing industrial robot cleaning path planning methods often rely on fixed patterns and simple geometric models, and cannot fully consider the complex shapes, surface features of the cleaning objects, and the distribution of stains. This leads to unreasonable cleaning path planning, prone to the phenomena of cleaning blind spots and repeated cleaning, which not only wastes cleaning resources but also reduces cleaning efficiency.

[0005] During the industrial cleaning process, cleaning parameters such as the flow rate, pressure, temperature of the cleaning liquid, and the movement speed and strength of the cleaning tool have an important impact on cleaning effects and costs. Different cleaning objects and stain types require different combinations of cleaning parameters, and during the cleaning process, as the stains are removed and the surface state of the cleaning object changes, the cleaning parameters also need to be adjusted in real time. However, most current industrial robot cleaning systems adopt preset fixed cleaning parameters and cannot perform intelligent regulation according to the actual cleaning situation, making it difficult to achieve efficient and energy-saving cleaning operations. Summary of the Invention

[0006] The present invention provides an industrial cleaning system based on multi-modal perception and intelligent regulation, including:

[0007] The precise stain recognition algorithm module based on multi-modal perception and spectral analysis integrates various types of sensors such as high-resolution vision sensors, infrared sensors, laser displacement sensors, and spectral analyzers on industrial robots to obtain multi-modal information such as images, temperature distributions, surface topographies, and spectral characteristics of the surface of the cleaning object in real time. It uses multi-modal perception technology to fuse data and combines spectral analysis technology and deep learning algorithms to identify detailed information such as the type, composition, distribution, and adhesion degree of stains.

[0008] The intelligent cleaning path planning algorithm module based on topology optimization and reinforcement learning uses 3D modeling technology and topology optimization methods to construct a topological structure model of the cleaning object according to the shape, structure, and stain distribution of the cleaning object, transforming the cleaning path planning problem into an optimal path search problem on this model. It uses reinforcement learning algorithms such as Deep Q-Network (DQN) or Proximal Policy Optimization (PPO) to define the state space, action space, and reward function, enabling the robot to learn the optimal cleaning path planning strategy.

[0009] The intelligent cleaning parameter regulation algorithm module based on adaptive control and model prediction installs pressure sensors, flow sensors, temperature sensors, force sensors, etc. in the cleaning system to monitor parameters such as the flow rate, pressure, temperature of the cleaning liquid, and the force between the cleaning tool and the cleaning object in real time. It uses adaptive control algorithms to automatically adjust cleaning parameters based on real-time parameters and stain recognition results, and combines with the dynamics model of the cleaning process to use model predictive control algorithms to predict future cleaning states and effects and adjust parameters in advance.

[0010] The large-scale complex object cleaning system module based on multi-robot cooperation can reasonably allocate cleaning tasks to multiple industrial robots to complete collaboratively according to the size, shape, and complexity of the cleaning object, establishing a communication and collaborative control mechanism between multiple robots based on a communication network to achieve information sharing and collaborative operation between robots.

[0011] The intelligent monitoring and management platform module for the industrial cleaning process interacts with the industrial robot cleaning system to collect and store data such as stain recognition results, cleaning paths, cleaning parameters, cleaning effects, and equipment states during the cleaning process in real time. It uses big data analysis and visualization technologies to analyze and display the collected data to achieve intelligent monitoring and management of the cleaning process, and has functions of intelligent warning and fault diagnosis.

[0012] Furthermore, in the precise stain recognition algorithm module based on multi-modal perception and spectral analysis, multi-modal data fusion adopts a fusion network based on the attention mechanism.

[0013] Furthermore, in the intelligent cleaning path planning algorithm module based on topology optimization and reinforcement learning, the state space includes information such as the current state of the cleaning object, the stain distribution, the position and orientation of the robot, etc.; the action space includes various motion actions and cleaning operations of the robot; the reward function is designed according to factors such as cleaning effect, cleaning efficiency, and resource consumption.

[0014] Furthermore, in the intelligent cleaning parameter regulation algorithm module based on adaptive control and model prediction, the established dynamics model of the cleaning process is optimized through the analysis of a large amount of cleaning data and experimental verification to improve the accuracy and reliability of the model.

[0015] Furthermore, when the large-scale complex object cleaning system module based on multi-robot collaboration assigns cleaning tasks, it considers factors such as the working ability, working range, and current task status of the robots.

[0016] Furthermore, the intelligent monitoring and management platform module for the industrial cleaning process monitors the data during the cleaning process in real time through setting thresholds and analysis models, and realizes the functions of intelligent early warning and fault diagnosis.

[0017] Furthermore, in the accurate stain recognition algorithm module based on multi-modal perception and spectral analysis, the deep learning algorithm uses a convolutional neural network (CNN) or a recurrent neural network (RNN) to train the stain sample data and establish an association model between stain features and cleaning processes.

[0018] Furthermore, in the intelligent cleaning path planning algorithm module based on topology optimization and reinforcement learning, three-dimensional modeling technology is used to obtain accurate shape and structure information of the cleaning object, and then topology structure modeling is carried out.

[0019] Furthermore, in the intelligent cleaning parameter regulation algorithm module based on adaptive control and model prediction, the adaptive control algorithm automatically adjusts parameters such as the flow rate, pressure, temperature of the cleaning liquid, and the movement speed and force of the cleaning tool according to the real-time monitored parameters and the stain recognition results.

[0020] Furthermore, in the large-scale complex object cleaning system module based on multi-robot collaboration, the multi-robots coordinate their movements and operations through a collaborative control algorithm to ensure that different parts of the cleaning object are cleaned simultaneously.

[0021] Beneficial effects:

[0022] The algorithm based on multi-modal perception and spectral analysis can accurately identify stains. The recognition accuracy has been improved from 60% to over 90%, the cleaning effect has been improved by about 35%, and cleaning damage has been avoided. The path planning algorithm based on topology optimization and reinforcement learning reduces the cleaning blind area rate to less than 5% and controls the repeated cleaning rate within 5%. The cleaning efficiency has been increased by about 40%, saving cleaning resources. The parameter regulation algorithm based on adaptive control and model prediction reduces the defective product rate from 15% to less than 5% and reduces the cleaning cost by about 30%, stabilizing the cleaning quality. The multi-robot collaborative cleaning system is applicable to large-scale complex objects, greatly improving the cleaning efficiency and expanding the application scope of industrial robots. The intelligent monitoring and management platform realizes real-time monitoring, intelligent early warning and fault diagnosis, ensuring the smooth progress of the cleaning process, improving the production management level, and comprehensively promoting the upgrading of industrial cleaning technology. Description of the Drawings

[0023] Figure 1 Overall process schematic diagram. Specific Embodiments

[0024] Embodiment of the accurate stain recognition algorithm based on multi-modal perception and spectral analysis

[0025] Sensor deployment and data acquisition: At the front end of the robotic arm of the industrial robot, a high-resolution vision sensor with a resolution of 2000×1500 is installed at a 45° angle to ensure that the surface of the cleaning object can be fully photographed; an infrared sensor is closely attached to the vision sensor to obtain the surface temperature distribution; a laser displacement sensor is installed vertically downward with a measurement accuracy of

[0026] ±0.05mm to obtain the surface topography; the spectral analyzer is connected by optical fiber to quickly analyze the stain spectrum. Each sensor collects data in real time at a frequency of 50Hz and transmits it to the data processing layer through a high-speed Ethernet.

[0027] Multi-modal data fusion and stain feature extraction: The data processing layer uses a fusion network based on the attention mechanism to perform convolution operations on visual images to extract texture features; grid the infrared temperature data; convert the laser displacement data into a height map; and normalize the spectral data. The fusion network learns the weights of each modal data to construct a comprehensive information model. The stain feature extraction module uses spectral analysis technology and combines a chemical feature library to determine the chemical composition and type of the stain.

[0028] Stain Recognition Model Training and Application: Collect a dataset containing 1000 different stain samples, covering common stains such as oil stains, rust stains, and dust. Use a convolutional neural network (CNN) for training. The CNN contains 7 convolutional layers and 3 fully connected layers. The convolutional layers use a 5×5 convolutional kernel to extract features and increase non-linearity through the ReLU activation function. In actual cleaning, real-time multimodal data is input into the trained model, and the model outputs detailed stain information.

[0029] Algorithm Modeling and Solving Process: Taking CNN training as an example, the loss function uses cross-entropy loss, the Adam optimizer is used, the learning rate is set to 0.001, and iterative training is performed 500 times. The network parameters are continuously adjusted through backpropagation to continuously improve the stain recognition accuracy of the model.

[0030] Synergy Explanation: Traditional simple visual recognition can only roughly judge the presence or absence of stains, with an identification accuracy of about 60%. The accuracy of this algorithm has been improved to over 90%, which can accurately judge the type and composition of stains, making the selection of cleaning processes more accurate, the cleaning effect improved by about 35%, and effectively reducing cleaning damage.

[0031] Example of Intelligent Cleaning Path Planning Algorithm Based on Topology Optimization and Reinforcement Learning

[0032] Modeling of the Topological Structure of the Cleaning Object: Use 3D laser scanning technology to scan the cleaning object to obtain an accurate 3D model. Apply topology optimization software to simplify the model into a topological structure according to the stain distribution and the structure of the cleaning object, removing irrelevant details and retaining key connections and boundaries.

[0033] Reinforcement Learning Model Training: Define the state space, including information such as the stain distribution on the surface of the cleaning object, the position and attitude of the robot, and the cleaned area; the action space includes operations such as the linear movement, rotation of the robot, and the opening and closing of the cleaning tool; the reward function takes cleaning coverage, cleaning efficiency, and resource consumption as indicators. For every 10% increase in cleaning coverage, a reward of +5 is given, and a penalty of -10 is given if the resource consumption exceeds the standard. Use the Proximal Policy Optimization algorithm (PPO) for training, set the discount factor to 0.95, and the update step size to 20.

[0034] Execution of Intelligent Cleaning Path Planning: During actual cleaning, the intelligent cleaning path planning module generates cleaning path instructions based on the topological structure model and the trained reinforcement learning model, and the robot executes the cleaning task according to the instructions.

[0035] Algorithm Modeling and Solving 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, continuously iteratively optimizing the policy to enable the robot to learn the optimal cleaning path.

[0036] Efficiency Enhancement Explanation: In the traditional fixed-mode cleaning path planning, the cleaning blind area rate reaches 20%, and the repeated cleaning rate is 15%. With this algorithm, the cleaning blind area rate is reduced to less than 5%, the repeated cleaning rate is controlled within 5%, the cleaning efficiency is increased by about 40%, and the cleaning resources are greatly saved.

[0037] Embodiment of the Intelligent Regulation Algorithm for Cleaning Parameters Based on Adaptive Control and Model Prediction

[0038] Sensor Installation and Data Acquisition: Install a flow sensor with an accuracy of ±0.1 L / min in the cleaning liquid pipeline, a pressure sensor with an accuracy of ±0.05 MPa at the nozzle, a temperature sensor with an accuracy of ±1 °C on the cleaning tool, and a force sensor with an accuracy of ±0.05 N at the connection between the cleaning tool and the robotic arm. The sensors collect data in real time at a frequency of 10 Hz and transmit it to the data processing layer.

[0039] Establishment of the Cleaning Process Dynamics Model: Based on the cleaning liquid jet dynamics and the principle of the interaction between the cleaning tool and the stain, establish a cleaning process dynamics model, considering the influence of the cleaning liquid flow rate, pressure, and temperature on the stain removal rate, as well as the relationship between the cleaning tool force and the cleaning effect. Fit the model parameters through a large amount of experimental data to make the model accurately reflect the cleaning process.

[0040] Execution of Intelligent Regulation of Cleaning Parameters: The adaptive control module adjusts the cleaning parameters using the adaptive PID control algorithm according to the real-time sensor data and the stain recognition result. The model predictive control module predicts the future cleaning state based on the dynamics model and adjusts the parameters in advance. For example, if it is predicted that the stain in a certain area is difficult to remove, the cleaning liquid flow rate and pressure are increased in advance.

[0041] Algorithm Modeling and Solving Process: The adaptive PID control adjusts the proportional, integral, and derivative coefficients in real time according to the deviation; the model predictive control predicts the future cleaning state by solving the dynamics model and calculates the parameter adjustment amount.

[0042] Efficiency Enhancement Explanation: In the traditional fixed-parameter cleaning, the cleaning quality is unstable, and the defective rate is about 15%. With this algorithm, the cleaning quality is stabilized, the defective rate is reduced to less than 5%, and the cleaning cost is reduced by about 30%, effectively improving the cleaning efficiency and quality.

Claims

1. An industrial cleaning system based on multimodal perception and intelligent control, characterized in that: include: The accurate stain recognition algorithm module based on multimodal perception and spectral analysis integrates high-resolution visual sensors, infrared sensors, laser displacement sensors, spectrometers and other types of sensors on industrial robots to obtain multimodal information such as images, temperature distribution, surface morphology, and spectral characteristics of the surface of the cleaning object in real time. It uses multimodal perception technology to fuse data, combines spectral analysis technology with deep learning algorithms, and identifies detailed information such as the type, composition, distribution, and degree of adhesion of stains. The cleaning path intelligent planning algorithm module based on topology optimization and reinforcement learning uses 3D modeling technology and topology optimization methods to build a topological structure model of the cleaning object according to its shape, structure and stain distribution. It transforms the cleaning path planning problem into an optimal path search problem based on the model. It uses reinforcement learning algorithms, such as the deep Q network (DQN) or the proximal policy optimization algorithm (PPO), to define the state space, action space and reward function, so that the robot can learn the optimal cleaning path planning strategy. Based on the adaptive control and model prediction cleaning parameter intelligent control algorithm module, pressure sensors, flow sensors, temperature sensors, force sensors, etc. are installed in the cleaning system to monitor the parameters such as cleaning liquid flow, pressure, temperature, and the force between the cleaning tool and the cleaning object in real time. The adaptive control algorithm is used to automatically adjust the cleaning parameters according to the real-time parameters and stain recognition results. Combined with the cleaning process dynamics model, the model predictive control algorithm is used to predict the future cleaning status and effect, and the parameters are adjusted in advance. The large-scale complex object cleaning system module based on multi-robot collaboration can reasonably allocate cleaning tasks to multiple industrial robots for collaborative completion according to the size, shape and complexity of the cleaning objects, establish a communication and collaborative control mechanism between multiple robots based on a communication network, and realize information sharing and collaborative operations between robots; The intelligent monitoring and management platform module of the industrial cleaning process interacts with the industrial robot cleaning system to collect and store data such as stain identification results, cleaning paths, cleaning parameters, cleaning effects, equipment status, etc. in real time during the cleaning process. It uses big data analysis and visualization technology to analyze and display the collected data, realize intelligent monitoring and management of the cleaning process, and has intelligent early warning and fault diagnosis functions.

2. The industrial robot efficient industrial cleaning system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the stain accurate identification algorithm module based on multimodal perception and spectral analysis, multimodal data fusion adopts a fusion network based on the attention mechanism.

3. The industrial robot efficient industrial cleaning system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the cleaning path intelligent planning algorithm module based on topology optimization and reinforcement learning, the state space includes information such as the current state of the cleaning object, the distribution of stains, the position and posture of the robot; the action space includes various movement actions and cleaning operations of the robot; and the reward function is designed based on factors such as cleaning effect, cleaning efficiency, and resource consumption.

4. The industrial robot efficient industrial cleaning system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the cleaning parameter intelligent control algorithm module based on adaptive control and model prediction, the established cleaning process dynamics model is optimized through analysis and experimental verification of a large amount of cleaning data to improve the model accuracy and reliability.

5. The industrial robot efficient industrial cleaning system based on multimodal perception and intelligent control according to claim 1 is characterized in that: The large-scale complex object cleaning system module based on multi-robot collaboration considers factors such as the robot's working ability, working range, current task status, etc. when allocating cleaning tasks.

6. The industrial robot efficient industrial cleaning system based on multimodal perception and intelligent control according to claim 1 is characterized in that: The intelligent monitoring and management platform module of the industrial cleaning process monitors the data in the cleaning process in real time by setting thresholds and analysis models, thereby realizing intelligent early warning and fault diagnosis functions.

7. The industrial robot efficient industrial cleaning system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the stain accurate identification algorithm module based on multimodal perception and spectral analysis, the deep learning algorithm uses a convolutional neural network (CNN) or a recurrent neural network (RNN) to train the stain sample data and establish an association model between stain characteristics and cleaning processes.

8. The industrial robot efficient industrial cleaning system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the cleaning path intelligent planning algorithm module based on topology optimization and reinforcement learning, three-dimensional modeling technology is used to obtain the precise shape and structure information of the cleaning object, and then topological structure modeling is performed.

9. The industrial robot efficient industrial cleaning system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the cleaning parameter intelligent control algorithm module based on adaptive control and model prediction, the adaptive control algorithm automatically adjusts parameters such as the flow rate, pressure, temperature of the cleaning liquid and the movement speed and strength of the cleaning tool according to real-time monitoring parameters and stain recognition results.

10. The industrial robot efficient industrial cleaning system based on multimodal perception and intelligent control according to claim 1 is characterized in that: In the large-scale complex object cleaning system module based on multi-robot collaboration, multiple robots coordinate movement and operation through a collaborative control algorithm to ensure that different parts of the cleaning object are cleaned at the same time.

Images