SBIR-STTR Award

Photonics Optics Tech (POT), Inc. proposes to develop an innovative Hybrid Neural Network Augmented Kalman Filter (HNN-KF) to enable flight vehicle trajectory estimate using TSPI data collected form coupled GPS-IMU sensor in noisy and dynamic testing environment encountered at AFFTC, and other EAFB ranges. The HNN-KF will be a software tool that will improve the inherent advantages of Kalman filter by the fusion of a self-learning RBF-NN algorithm. The HNN-KF algorithm, upon completion of development, will be inserted into the SOA MOSES software to replace the current Kalman filter algorithm to conduct trajectory estimate using GPS-IMU TSPI data provided by EAFB. The adaptive self-learning HNN-KF algorithm will provide automation in trajectory estimate computation without having to operate in the conventional Man in the Loop mode that will result in greatly improve trajectory estimate accuracy as well as the significant reduction in TSPI processing time as well as the reduction of man hours. This will also result in great cost reduction. Upon successful feasibility demonstration and validation, POT will propose in Phase II work plan to develop a full-fledged, hardware-in-the-loop HNN-KF intelligent algorithm and corresponding source coding for real-time high-performance flight vehicle trajectory estimate

Benefits:
Primary benefits of the proposed Hybrid Kalman filter-Neural network algorithm development will improve flight vehicle trajectory estimate with 1) improved accuracy, 2) processing automation by eliminate "Man-in-the-loop" process and 3) great cost reduction due to labor and time savings. The HKF-NN algorithms will be a powerful and versatile target tracking, trajectory estimate tool that will benefit all types of auto-piloted flight vehicles for both military and civilian applications.

Keywords:
Gps, Imu, State Parameter, Estimate Of Trajectory, Algorithm, Neural Network, Kalman Filter, Data Fusion

The scope of the proposed Phase II task is to develop an intelligent MOSES system that is capable of autonomous updating parameter file of the KF within the MOSES. This will enable the dynamic updating and reduction of KF errors during the flight vehicle trajectory post-processing. The proposed work develop is based on the successful Phase I preliminary Kalman Filter Neural Network algorithms design and the trajectory reconstruction simulation study that have proved the soundness of the proposed approach using a trained NN to improve the KF performance. The Phase II R&D effort will include the development of 1) A Clustering Ensemble Approached based Neural Network that will generate input to the MOSES parameter files updating. Extensive experimental studies will be performed using the intelligent MOSES system with real flight trajectory data to demonstrate its performance capability.

Benefit:
Upon completion of the development of the advanced intelligent MOSES system that is capable of autonomous updating parameter file of the KF within the MOSES. This intelligent MOSES software will enable the dynamic updating and reduction of KF errors during the flight vehicle trajectory post-processing.The real-time update capability will benefit the MOSES operations in at least two ways: 1. Shorten the trajectory post-processing time using the SOA MOSES due to the elimination of the “human-in-the-loop” parameters tuning work. 2. Improve the trajectory reconstruction accuracy due to the in-time update of the KF parameter files.

Keywords:
Kalman Filter, Extended Kalman Filter, Moses, Neural Network, Clustering Ensemble Approach, Trajectory Post-Processing