Neural Network Position Determination for Irregular Capacitive Sensors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional position determination methods for capacitive cell structures, such as touchscreens and liquid level gauges, fail to accurately and efficiently handle non-planar, curve-shaped, or irregularly shaped surfaces, resulting in unreliable results and performance issues like accuracy, latency, and memory footprint problems.
Innovation Solution
The implementation of machine learning-based methods using trainable neural networks to adapt to the geometry of capacitive cell structures, allowing for reliable position determination by adjusting edge weights and network parameters through supervised and unsupervised training, and employing techniques like pruning and quantization to improve computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional position determination methods are used for capacitive cell structures, then the system is simple to implement, but it fails to accurately handle non-planar, curve-shaped, or irregularly shaped surfaces resulting in unreliable results
Solution Approach 1:
The patent transforms the position determination problem from geometric parameter-based calculations to machine learning parameter-based predictions. By training neural networks with capacitance values and corresponding position labels, the system learns optimal parameters for accurately determining positions on surfaces of any shape, resolving the contradiction between reliability and complexity.
Solution Approach 2:
The patent replaces conventional geometric and mathematical calculation methods with machine learning-based prediction systems. Neural networks substitute traditional algorithms, enabling accurate position determination on complex surfaces without requiring explicit geometric models, thus improving reliability while managing system complexity.
2Loss of time
If conventional position determination methods are used, then computational resources are conserved, but latency and memory footprint problems occur when handling complex surfaces
Solution Approach 1:
The patent applies preliminary action by training neural networks offline before deployment. During the training phase, the system processes large datasets to learn position patterns, storing learned parameters in the network weights. During actual position determination, the pre-trained network rapidly predicts positions without requiring complex real-time calculations, significantly reducing latency while maintaining processing efficiency.
3Measurement precision
If machine learning-based methods with trained neural networks are employed, then position determination accuracy is improved, but memory footprint increases
Solution Approach 1:
The patent manages memory footprint by optimizing neural network parameters through pruning and quantization techniques. By removing redundant connections and reducing precision requirements, the system maintains high position determination accuracy while significantly reducing the memory resources required to store network weights and structures.
4Adaptability or versatility
If conventional methods assume planar rectangular shape, then the system is easy to manufacture, but it cannot adapt to non-planar, curve-shaped, or irregularly shaped surfaces
Solution Approach 1:
The patent achieves universality by developing a single machine learning-based position determination system that can handle surfaces of any shape. The neural network learns from training data representing various surface geometries, enabling the same system to accurately determine positions on planar, curved, irregular, and non-compact surfaces without requiring shape-specific algorithms, thus improving adaptability while maintaining ease of manufacture.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach provides accurate and efficient position determination with reduced dependence on the sensor's form factor, improving performance in terms of accuracy, latency, and memory usage, and reducing development time and effort.
Implementation Method 1
Bringing a finger or a conductive stylus to the surface of a capacitive cell would change its capacitance, which may be measured and then utilized for touch position determination.
Data Source
AI summary
An example method of determining the position value reflecting an action applied to the capacitive sensor device comprises: receive a set of capacitance values of a plurality of capacitive cells of a capacitive sensor device; determining a local maximum of the set of capacitance values; identifying a set of neural network parameters corresponding to the local maximum of the set of capacitance values; and processing the set of capacitance values by a neural network using the identified set of neural network parameters to determine a position value reflecting an action applied to the capacitive sensor device.