Signal Processing Device Using Stacked Autoencoder for Centrifugal Force Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing signal processing technologies face challenges in efficiently extracting feature quantities for specific events like centrifugal force from sensor signals, particularly in embedded systems, due to high computational requirements and the difficulty in designing correction filters.
Innovation Solution
A signal processing device incorporating a stacked autoencoder, a control line associated learner, and a refactorer to extract feature quantities by processing input signals from sensors, using deep learning to differentiate between event aspects and adjust extracted features, allowing for efficient removal of unwanted components like centrifugal force from motion or image data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a convolutional neural network (CNN) is used for feature quantity extraction, then the feature extraction capability is improved, but the calculation resources required increase significantly
Solution Approach 1:
The patent segments the feature extraction task by using a stacked autoencoder to first extract basic features from sensor signals, then using a control line associated learner to specifically extract event-related features (such as centrifugal force). This segmentation allows the system to achieve accurate feature extraction without requiring the full computational power of a CNN, as each component handles a specific aspect of feature extraction.
Solution Approach 2:
The patent extracts only the necessary features for specific events (like centrifugal force detection) rather than performing comprehensive image-like analysis. The control line associated learner is designed to extract only the relevant event aspects from the preprocessed features, eliminating unnecessary computational overhead while maintaining extraction accuracy for the target events.
2Adaptability or versatility
If deep learning technology is used to extract feature quantities from sensor signals, then the ability to handle events difficult to formulate mathematically is improved, but the device complexity increases
Solution Approach 1:
The patent applies local quality by designing the control line associated learner to focus on specific event aspects (such as centrifugal force or focus) rather than attempting to analyze all possible signal characteristics. Each control line is associated with a specific event aspect, allowing the system to handle complex events with targeted analysis rather than comprehensive processing, thereby reducing overall system complexity.
Solution Approach 2:
The patent performs preliminary action by using the stacked autoencoder to preprocess sensor signals and extract basic features before feeding them to the control line associated learner. This preliminary processing simplifies the input data structure and reduces the complexity of subsequent event-specific analysis, making the overall system more manageable while maintaining high adaptability.
3Measurement precision
If correction filters are designed to remove unwanted components like centrifugal force, then the signal accuracy is improved, but the ease of manufacture deteriorates due to difficulty in creating mathematical models
Solution Approach 1:
The patent applies self-service by using the control line associated learner to automatically learn and adapt to the characteristics of unwanted components (such as centrifugal force) from the sensor signals. Instead of requiring manual mathematical model creation and filter design, the system automatically identifies and removes these components through the learning process, significantly easing the manufacturing and deployment process while maintaining high signal accuracy.
Solution Approach 2:
The patent uses parameter changes by adjusting the control line parameters in the associated learner to optimize the removal of unwanted components. The system learns optimal parameter values during the training phase, allowing it to adapt to different signal characteristics and unwanted component patterns without requiring manual filter redesign, thus improving both accuracy and ease of implementation.
Data Source
AI summary
A signal processing device according to the present technology includes a stacked auto encoder that processes an input signal from a sensor, a control line associated learner including a neural network and subjected to control line associated learning for performing learning by associating different event aspects related to a specific event with values of different control lines with a feature quantity obtained in an intermediate layer of the stacked auto encoder after pretraining as an input, and a refactorer that obtains a difference between a first output that is an output of the control line associated learner when a first value is given to the control line, and a second output that is an output of the control line associated learner when a second value different from the first value is given to the control line.