Machine Motion Control Using CNN-LSTM Sensor Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-performance positioning technologies for manufacturing and metrology achieve submicron precision but are costly, limiting their practical applications in industrial processes.
Innovation Solution
A control system utilizing a deep learning model comprising a convolutional neural network (CNN) and a long short-term memory (LSTM) that integrates data from different types of sensors to predict and correct machine motion, allowing for precise control using low-cost sensor arrays and actuators.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If high-performance positioning technologies (air-bearing stages, flexure-based piezoelectric nanopositioners) are used to achieve submicron precision, then manufacturing precision and repeatability are improved, but cost increases significantly
Solution Approach 1:
The patent replaces expensive, high-performance sensors and actuators with low-cost sensor arrays and standard actuators. The system uses multiple inexpensive sensors to capture motion information from multiple perspectives, compensating for individual sensor limitations through data fusion and deep learning processing, thereby achieving high precision without relying on costly components
Solution Approach 2:
The patent substitutes traditional mechanical precision engineering approaches with an information-processing approach. Instead of relying on precision-machined mechanical components and high-performance sensors, the system uses standard components combined with deep learning models (CNN and LSTM) to process sensor data and achieve submicron positioning precision through intelligent algorithms
2Measurement precision
If deep learning models are trained using data from high-cost sensors, then model accuracy and control precision are improved, but the initial training cost and complexity increase
Solution Approach 1:
The patent performs preliminary training of the deep learning model using high-cost sensor data to establish an accurate reference model. Once trained, the model can operate with low-cost sensors, effectively performing the expensive calibration phase once and then using inexpensive components for ongoing operation. This preliminary action transfers the complexity burden to an initial setup phase rather than continuous operation
Solution Approach 2:
The patent introduces high-cost sensor data as an intermediary training resource that mediates between the available low-cost sensors and the desired high precision output. The training data acts as a bridge, allowing the model to learn from accurate references and then apply that knowledge to compensate for the limitations of inexpensive sensors during actual operation
Data Source
AI summary
In some embodiments, a control system, a control method and a storage medium are provided. In the method, first motion information of a machine acquired by a first sensor is received; the first motion information is inputted into a deep learning model to obtain a model output, the deep learning model comprising a convolutional neural network (CNN) and a long short-term memory (LSTM); the deep learning model is trained using the first motion information and second motion information acquired by a second sensor; the first sensor and the second sensor having different ways of detecting information and processing the detected information. The model output is used to control the machine.