Hybrid Vibration Prediction Model for Non-Linear Load Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing time series data predicting devices, such as those using regression neural networks, struggle to accurately predict the behavior of vibration proofing members with non-linear characteristics due to the inability to properly consider fluctuating data over time, leading to deteriorated prediction accuracy.
Innovation Solution
A predicting device that separates the modeling of linear and non-linear characteristics of a vibration proofing member using a dynamic system model and a regression neural network, respectively, to generate load data by combining first and second load data inferred from displacement and velocity data, thereby improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a regression neural network is used to predict the behavior of a vibration proofing member with non-linear characteristics, then the model can handle non-linear data, but the prediction accuracy deteriorates because the model cannot properly consider fluctuating data over time
Solution Approach 1:
The patent segments the prediction model into two distinct components: a dynamic system model that handles time-series fluctuating data and a regression neural network that handles non-linear characteristics. This segmentation allows each model to specialize in its strength, resolving the contradiction between handling non-linearity and maintaining prediction accuracy.
Solution Approach 2:
The patent merges the dynamic system model and the regression neural network into a hybrid prediction model. The dynamic system model processes displacement and velocity data to capture time-dependent behavior, while the regression neural network processes the same data to capture non-linear characteristics. The outputs of both models are combined to produce the final prediction, thereby achieving both non-linear handling and high prediction accuracy.
2Loss of substance
If only compressed data is used to generate a prediction model, then storage space is optimized, but the prediction accuracy is not considered and deteriorates
Solution Approach 1:
Instead of using only compressed data, the patent uses both the original displacement data and velocity data in full, along with their compressed representations. This partial or excessive use of data ensures that sufficient information is retained for accurate prediction while still benefiting from compression where appropriate.
Solution Approach 2:
The patent introduces an intermediary processing step where both compressed and original data are fed into the hybrid model. The dynamic system model and regression neural network act as intermediaries that process the data in different ways, ensuring that neither data compression nor data fidelity is sacrificed, and both contribute to the final accurate prediction.
Data Source
AI summary
A predicting device, including a processor configured to: acquire displacement data that expresses a time series of displacements at respective points in time that are input to a vibration proofing member, and velocity data that expresses a time series of velocities at respective points in time that are input to the vibration proofing member; generate first load data of the vibration proofing member by inputting the acquired displacement data and velocity data into a model that is for inferring, from the displacement data and the velocity data, load data; generate second load data of the vibration proofing member by inputting the acquired displacement data and velocity data into a regression trained model that is for inferring, from the displacement data and the velocity data, load data; and infer load data relating to the vibration proofing member by adding together the generated first load data and the generated second load data.