Weigh-in-Motion System Based on an Improved Kalman and LSTM-Attention Algorithm.

Area of Science:

• Engineering
• Computer Science
• Logistics Technology

Background:

• Weigh-in-motion (WIM) systems are crucial for logistics and industry, but accuracy is challenged by shock and vibration.
• Increasing labor and time costs necessitate more efficient and precise WIM solutions.

Purpose of the Study:

• To develop a robust algorithm for filtering noise in WIM systems.
• To create an accurate deep learning model for predicting object mass using sensor data.

Main Methods:

• A novel six degrees-of-freedom (DOF), mass-spring damping-based Kalman filter with time scale (KFTS) algorithm was designed for noise filtering.
• An attention-based long short-term memory (LSTM) model was developed to predict mass from time-series sensor signals.

Main Results:

• The KFTS algorithm effectively filters noise coupled with sensor signals.
• The attention-based LSTM model demonstrated superior performance over SVM, FCN, and XGBoost models.
• The deep learning model achieved remarkable accuracy across various loads, speeds, and operating conditions.

Conclusions:

• The proposed advanced Kalman filter and LSTM model significantly improve WIM system accuracy and stability.
• This deep learning approach offers a viable solution for high-precision applications in the logistics industry.