Adaptive chip mounter path optimization method based on artificial intelligence
Through the adaptive path optimization method based on artificial intelligence, the problem of large amount of computation and inefficiency of path planning of patch machines is solved, and more efficient, stable and accurate path planning is achieved to meet the production needs of high density and complexity.
Patent Information
- Application Number
- CN202411251626.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-08
- Publication Date
- 2024-12-13
AI Technical Summary
The existing patch machine has a large amount of calculation, low efficiency, and insufficient path optimization, making it difficult to meet the production needs of high density, complexity and flexibility.
Adaptive path optimization method based on artificial intelligence is adopted to obtain initial state information, state representation transformation, design complex reward functions, automatic feature extraction, multi-objective reward function design, introduction of sparse and stability reward functions, and consider path planning Time factor, deep learning models and multi-objective optimization algorithms are used for path planning.
It improves the accuracy and accuracy of path planning, enhances the stability and robustness of the system, ensures that the path planning completes tasks within the specified time, and improves the operating efficiency and production efficiency of the patch machine.
Smart Images
Figure CN119136416A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of surface mounting technology of chip mounters, and specifically to an adaptive chip mounter path optimization method based on artificial intelligence. Background Art
[0002] With the rapid development of the electronics manufacturing industry, the production demand for printed circuit boards is getting higher and higher, and it needs to meet the requirements of high density, high complexity, large scale and flexibility. As the mainstream technology for printed circuit board production, surface mount technology has been rapidly developed. The application of chip mounters in the production process is becoming more and more extensive. Chip mounters are widely used in the production of electronic products. They can efficiently complete the placement of electronic components. Chip mounter path optimization is a key link to improve production efficiency and reduce costs.
[0003] In the prior art, the path planning of the placement machine mainly relies on traditional algorithms, but these algorithms have problems such as large amount of calculation, low efficiency, insufficient path optimization, etc. when processing complex scenes. In order to solve these problems, the present invention proposes an adaptive placement machine path optimization method based on artificial intelligence. Summary of the invention
[0004] In view of the deficiencies in the prior art, the present invention provides an adaptive placement machine path optimization method based on artificial intelligence, which solves the problems of large computational complexity, low efficiency and insufficient path optimization in the path planning of the existing placement machines.
[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an adaptive placement machine path optimization method based on artificial intelligence, comprising the following steps:
[0006] S1, obtain the initial state information of the placement machine;
[0007] S2, state representation transition;
[0008] S3, design complex reward functions;
[0009] S4, automatic feature extraction method;
[0010] S5, design multi-objective reward function;
[0011] S6, introduce sparse reward function;
[0012] S7, design stability reward function;
[0013] S8, consider the time factor of path planning;
[0014] S9, reward function combination.
[0015] Preferably, the step of obtaining the initial status information of the placement machine includes:
[0016] S1-1, using encoder sensors to detect the position, speed and direction of the placement machine;
[0017] S1-2, sensor data includes position coordinates (x, y), linear velocity, angular velocity and other information;
[0018] S1-3, the data is converted into digital signals through an analog-to-digital converter and transmitted to the control system.
[0019] Preferably, the state representation conversion includes:
[0020] S2-1, uses convolutional neural network (CNN) to convert raw sensor data into feature maps to capture the spatial information of the current state of the placement machine;
[0021] S2-2, using recurrent neural network (RNN) to process time series data and capture the temporal dependencies between states;
[0022] S2-3, the neural network is trained through the back-propagation algorithm to optimize the network parameters to extract the most relevant feature representations.
[0023] Preferably, the complex reward function includes:
[0024] S3-1, Path length reward: Positive rewards are given according to the degree of reduction in path length. For every 1 meter reduction in path length, the reward increases by 0.2 points;
[0025] S3-2, Energy Consumption Reward: Positive rewards are given according to the degree of reduction in energy consumption. For every 0.1 unit reduction in energy consumption, the reward increases by 0.5 points;
[0026] S3-3, Production efficiency reward: Positive rewards are given according to the degree of improvement in production efficiency. For every 1% increase in production efficiency, the reward increases by 10 points.
[0027] Preferably, the automatic feature extraction method comprises:
[0028] S4-1, using autoencoder to learn latent feature representations in data;
[0029] S4-2, the autoencoder consists of an encoder and a decoder, which learns a compressed representation of the data by minimizing the reconstruction error;
[0030] S4-3, the learned feature representation can help the model better understand the data and improve the accuracy and generalization ability of path planning.
[0031] Preferably, the design of the multi-objective reward function includes:
[0032] S5-1, comprehensively considers multiple objectives such as path length, energy consumption and production efficiency;
[0033] S5-2, setting the weights between different objectives can adjust the model's preference when optimizing path planning. The objective weights are path length (0.3), energy consumption (0.5), and production efficiency (0.2);
[0034] S5-5, optimize through the multi-objective optimization algorithm NSGA-II to find a balanced solution.
[0035] Preferably, the introducing a sparse reward function comprises:
[0036] S6-1, the reward is increased by 50 points when the key task is completed, and the reward at other time points is 0.1 point.
[0037] Preferably, the design stability reward function includes:
[0038] S7-1, the stability reward function penalizes unstable path planning behavior, and the penalty is reduced by 20 points for each sharp turn or sudden stop.
[0039] Preferably, the time factor considered in path planning includes:
[0040] S8-1, time limit: the task must be completed within 300 seconds;
[0041] S8-2, timeout penalty: for every second exceeding the time limit, the penalty is reduced by 5 points.
[0042] Preferably, the reward function combination includes:
[0043] S9-1, combined reward function: path length reward + energy consumption reward + production efficiency reward + sparse reward + time factor reward;
[0044] S9-2, combined reward function weights: path length (0.2), energy consumption (0.3), production efficiency (0.2), sparse reward (0.1), time factor (0.2).
[0045] The present invention provides an artificial intelligence-based adaptive placement machine path optimization method, which has the following beneficial effects:
[0046] 1. Through the deep learning model and automatic feature extraction method, the system of the present invention can more accurately capture the state information of the placement machine, thereby improving the precision and accuracy of path planning. By using the multi-objective optimization algorithm and complex reward function design, the system can comprehensively consider multiple goals and perform effective path planning, thereby improving the operation efficiency and production efficiency of the placement machine.
[0047] 2. The present invention designs a stability reward function and a sparse reward function. The system can punish unstable behaviors and reward the completion of key tasks, thereby enhancing the stability and robustness of the system. Considering the time factor of path planning and the timeout penalty mechanism, the system can complete the task within the specified time and punish the timeout behavior in time, thereby ensuring the real-time and reliability of path planning. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 It is a flow chart of the steps of the present invention. DETAILED DESCRIPTION
[0049] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0050] Example:
[0051] Please refer to the attached Figure 1 The embodiment of the present invention provides an adaptive placement machine path optimization method based on artificial intelligence, comprising the following steps:
[0052] S1, obtain the initial state information of the placement machine;
[0053] S2, state representation transition;
[0054] S3, design complex reward functions;
[0055] S4, automatic feature extraction method;
[0056] S5, design multi-objective reward function;
[0057] S6, introduce sparse reward function;
[0058] S7, design stability reward function;
[0059] S8, consider the time factor of path planning;
[0060] S9, Reward function combination.
[0061] The obtaining of the initial state information of the placement machine includes:
[0062] S1-1, using encoder sensors to detect the position, speed and direction of the placement machine. By using encoder sensors to detect the position, speed and direction of the placement machine, the system can obtain the movement status information of the placement machine in real time;
[0063] S1-2, the sensor data includes position coordinates (x, y), linear velocity, angular velocity and other information, wherein the position coordinates (x, y) represent the position information of the placement machine on the plane, and the linear velocity and angular velocity represent the movement speed and turning speed of the placement machine;
[0064] S1-3, the data is converted into digital signals through an analog-to-digital converter and transmitted to the control system, which converts the analog signals collected by the sensor into digital signals for subsequent calculation and processing.
[0065] The state representation transition includes:
[0066] S2-1, using convolutional neural network (CNN) to convert raw sensor data into feature maps, capturing the spatial information of the current state of the placement machine, and extracting features from the sensor data through CNN to obtain the spatial features of the placement machine state;
[0067] S2-2, using recurrent neural network (RNN) to process time series data, capturing the temporal dependency between states, and processing the spatiotemporal features of sensor data through RNN to obtain the temporal dependency of the placement machine state;
[0068] S2-3, the neural network is trained through the back-propagation algorithm to optimize the network parameters to extract the most relevant feature representations. By continuously adjusting the network weights and biases, the model can better extract the characteristics of the placement machine status.
[0069] The complex reward function includes:
[0070] S3-1, Path length reward: Positive rewards are given according to the degree of reduction in path length. For every 1 meter of path length reduction, the reward increases by 0.2 points, which encourages the system to find a more efficient path and reduce the travel distance, thereby saving time and improving production efficiency. For every 1 meter of path length reduction, the reward increases by 0.2 points. This design can encourage the system to choose a shorter path as much as possible, reduce the travel distance of the placement machine, and improve production efficiency;
[0071] S3-2, Energy consumption reward: Give positive rewards according to the degree of energy consumption reduction. For every 0.1 unit of energy consumption reduction, the reward increases by 0.5 points, encouraging the system to reduce energy consumption and improve energy efficiency, thereby reducing production costs and environmental impact. For every 0.1 unit of energy consumption reduction, the reward increases by 0.5 points. This design can encourage the system to consider energy efficiency in path planning and choose a more energy-efficient path plan;
[0072] S3-3, Production efficiency reward: Positive rewards are given according to the degree of improvement in production efficiency. For every 1% increase in production efficiency, the reward increases by 10 points, which encourages the system to improve production efficiency, achieve faster production cycles and higher output, thereby improving overall production performance. For every 1% increase in production efficiency, the reward increases by 10 points. This design can prompt the system to optimize path planning to maximize production efficiency, improve production capacity and competitiveness.
[0073] The automatic feature extraction method comprises:
[0074] S4-1, using autoencoders to learn the potential feature representations in the data, and using autoencoders to reduce the dimension and extract features of sensor data. Autoencoders are an unsupervised learning method that can help the system learn the potential feature representations in the data, thereby achieving data dimensionality reduction and feature extraction. By using autoencoders to reduce the dimension and extract features of sensor data, the system can extract the most representative and useful features from the raw data, providing more effective input for subsequent path planning and control;
[0075] S4-2, the autoencoder consists of an encoder and a decoder, which learns the compressed representation of data by minimizing the reconstruction error and realizes feature extraction of sensor data. The encoder is responsible for mapping the input data to the latent feature space, and the decoder is responsible for mapping the latent features back to the original data space, and learning the compressed representation of data by minimizing the reconstruction error. Through the structure of the encoder and decoder, the autoencoder can effectively learn the compressed representation of data and help the system discover important features and patterns in the data;
[0076] S4-3, the learned feature representation can help the model better understand the data and improve the accuracy and generalization of path planning. Through the learned feature representation, the system can better understand the sensor data, thereby improving the accuracy and generalization of path planning and control systems. The learned feature representation can help the system better capture the key features and patterns in the data and improve the system's ability to recognize the environment, thereby achieving more intelligent and efficient path planning and control operations.
[0077] The design of the multi-objective reward function includes:
[0078] S5-1, comprehensive consideration of multiple goals such as path length, energy consumption and production efficiency. By comprehensively considering multiple goals such as path length, energy consumption and production efficiency, the system can balance the relationship between different goals in the path planning process and achieve multi-faceted optimization. Considering multiple goals can enable the system to more comprehensively consider different needs and restrictions in path planning, thereby finding a more comprehensive and optimized solution;
[0079] S5-2, setting the weights between different goals can adjust the model's preference when optimizing path planning. The target weights are path length (0.3), energy consumption (0.5), and production efficiency (0.2). By setting the weights between different goals, the model's preference when optimizing path planning can be adjusted, thereby affecting the final optimization result. In the example, the weight of path length is 0.3, the weight of energy consumption is 0.5, and the weight of production efficiency is 0.2. This design can guide the system to pay more attention to the optimization of energy consumption, while also considering the impact of path length and production efficiency;
[0080] S5-3, optimize through the multi-objective optimization algorithm NSGA-II to find a balanced solution. NSGA-II is a commonly used multi-objective optimization algorithm that can help the system find a balanced solution while considering the trade-off relationship between multiple objectives. The NSGA-II algorithm can effectively search for a set of non-dominated solutions, which do not have obvious advantages and disadvantages between various objectives, thereby providing decision makers with diverse choices and helping the system to achieve optimization of multi-objective path planning.
[0081] The introduction of the sparse reward function includes:
[0082] S6-1, when the key task is completed, the reward is increased by 50 points, and the reward at other time points is 0.1 points, which motivates the system to invest more resources in key tasks. By increasing the reward by 50 points when the key task is completed, the system can be motivated to invest more resources and efforts in key tasks, thereby increasing the system's attention to key tasks. Introducing this sparse reward function can help the system to solve key tasks more focused and effectively, and improve the system's task execution efficiency and accuracy.
[0083] The design stability reward function includes:
[0084] S7-1, the stability reward function punishes unstable path planning behavior. For each sharp turn or emergency stop, the penalty is reduced by 20 points, which prompts the system to plan a more stable path. By punishing each sharp turn or emergency stop, reducing the penalty by 20 points, the system can plan a more stable path and avoid unstable behavior. This stability reward function can help the system plan a more stable and reliable path, and improve the safety and stability of the system in actual operation;
[0085] The time factors considered in path planning include:
[0086] S8-1, time limit: The task must be completed within 300 seconds. Setting a time limit for the task to be completed within 300 seconds can help the system to efficiently complete the path planning task within the specified time and improve the sense of urgency and efficiency of task execution. Setting a time limit can prompt the system to make the best path planning decision within a limited time and ensure that the task is completed on time;
[0087] S8-2, Overtime penalty: For every 1 second overtime, the penalty is reduced by 5 points to ensure that the system completes the task within the specified time. By setting up an overtime penalty mechanism, that is, reducing the penalty by 5 points for every 1 second overtime, the system can be ensured to complete the task within the specified time and avoid unlimited extension of path planning time. Overtime penalties can motivate the system to use time efficiently, avoid wasting time and resources, and improve the system's time management capabilities and task execution efficiency.
[0088] The reward function combination includes:
[0089] S9-1, combined reward function: path length reward + energy consumption reward + production efficiency reward + sparse reward + time factor reward, path length reward: encourages the system to choose the shortest path, thereby reducing driving distance and time and improving efficiency. Energy consumption reward: encourages the system to choose energy-efficient paths, reduce energy consumption, help save resources and reduce costs. Production efficiency reward: the reward system considers production efficiency factors, helps improve task execution efficiency, and ensures that tasks are completed on time. Sparse reward: encourages the system to invest more resources in key tasks or areas to ensure that the system attaches importance to key tasks. Time factor reward: ensures that the system completes tasks within the specified time, improves task execution efficiency, and avoids delays and waste of time;
[0090] S9-2, combined reward function weights: path length (0.2), energy consumption (0.3), production efficiency (0.2), sparse reward (0.1), time factor (0.2), by combining these reward functions, the system can comprehensively consider multiple key factors such as path length, energy consumption, production efficiency, task importance and time factor in the path planning process, thereby helping the system to make more comprehensive and reasonable path planning decisions. Adjusting the weights of different reward functions can balance the importance of various factors, allowing the system to respond more intelligently to path planning requirements in different situations and improve the system's adaptability and performance.
[0091] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. An adaptive placement machine path optimization method based on artificial intelligence, characterized in that: The method comprises the following steps: S1, obtain the initial state information of the placement machine; S2, state representation transition; S3, design complex reward functions; S4, automatic feature extraction method; S5, design multi-objective reward function; S6, introduce sparse reward function; S7, design stability reward function; S8, consider the time factor of path planning; S9, reward function combination.
2. The method for adaptive placement machine path optimization based on artificial intelligence according to claim 1, characterized in that: The obtaining of the initial state information of the placement machine includes: S1-1, using encoder sensors to detect the position, speed and direction of the placement machine; S1-2, sensor data includes position coordinates (x, y), linear velocity, angular velocity and other information; S1-3, the data is converted into digital signals through an analog-to-digital converter and transmitted to the control system.
3. The method for adaptive placement machine path optimization based on artificial intelligence according to claim 1, characterized in that: The state representation transition includes: S2-1, uses convolutional neural network (CNN) to convert raw sensor data into feature maps to capture the spatial information of the current state of the placement machine; S2-2, using recurrent neural network (RNN) to process time series data and capture the temporal dependencies between states; S2-3, the neural network is trained through the back-propagation algorithm to optimize the network parameters to extract the most relevant feature representations.
4. The method for adaptive placement machine path optimization based on artificial intelligence according to claim 1, characterized in that: The complex reward function includes: S3-1, Path length reward: Positive rewards are given according to the degree of reduction in path length. For every 1 meter reduction in path length, the reward increases by 0.2 points; S3-2, Energy Consumption Reward: Positive rewards are given according to the degree of reduction in energy consumption. For every 0.1 unit reduction in energy consumption, the reward increases by 0.5 points; S3-3, Production efficiency reward: Positive rewards are given according to the degree of improvement in production efficiency. For every 1% increase in production efficiency, the reward increases by 10 points.
5. The method for adaptive placement machine path optimization based on artificial intelligence according to claim 1, characterized in that: The automatic feature extraction method comprises: S4-1, using autoencoder to learn latent feature representations in data; S4-2, the autoencoder consists of an encoder and a decoder, which learns a compressed representation of the data by minimizing the reconstruction error; S4-3, the learned feature representation can help the model better understand the data and improve the accuracy and generalization ability of path planning.
6. The method for adaptive placement machine path optimization based on artificial intelligence according to claim 1, characterized in that: The design of the multi-objective reward function includes: S5-1, comprehensively considers multiple objectives such as path length, energy consumption and production efficiency; S5-2, setting the weights between different objectives can adjust the model's preference when optimizing path planning. The objective weights are path length (0.3), energy consumption (0.5), and production efficiency (0.2); S5-5, optimize through the multi-objective optimization algorithm NSGA-II to find a balanced solution.
7. The method for adaptive placement machine path optimization based on artificial intelligence according to claim 1 is characterized in that: The introduction of the sparse reward function includes: S6-1, the reward is increased by 50 points when the key task is completed, and the reward at other time points is 0.1 point.
8. The method for adaptive placement machine path optimization based on artificial intelligence according to claim 1 is characterized in that: The design stability reward function includes: S7-1, the stability reward function penalizes unstable path planning behavior, and the penalty is reduced by 20 points for each sharp turn or sudden stop.
9. The method for adaptive placement machine path optimization based on artificial intelligence according to claim 1, characterized in that: The time factors considered in path planning include: S8-1, time limit: the task must be completed within 300 seconds; S8-2, timeout penalty: for every second exceeding the time limit, the penalty is reduced by 5 points.
10. The method for adaptive placement machine path optimization based on artificial intelligence according to claim 1, characterized in that: The reward function combination includes: S9-1, combined reward function: path length reward + energy consumption reward + production efficiency reward + sparse reward + time factor reward; S9-2, combined reward function weights: path length (0.2), energy consumption (0.3), production efficiency (0.2), sparse reward (0.1), time factor (0.2).