Lebesgue-Sampling-based Deep Learning for Battery Diagnosis and Prognosis

The University of South Carolina is offering licensing opportunities for Lebesgue-Sampling-based Deep Learning for Battery Diagnosis and Prognosis

Website

http://techfinder.sc.edu/technology/48817

Background:

Accurate and efficient modeling of battery degradation is of great challenge and is becoming more and more complex for batteries in modern applications. Traditional degradation modeling methods have difficulty capturing these characteristics, and the FPD performance will be affected. Deep learning algorithms have powerful feature extraction and learning abilities. However, they are not capable of uncertainty management, which is critical in FDP, especially long-term prediction in prognosis.

Invention Description:

This patent proposed a Lebesgue Sampling fault diagnosis and prognosis (LS-FDP) framework that integrates deep learning techniques’ automatic learning capability and Bayesian estimation’s uncertainty management capability to address their limitations. With the integration of DBN for data processing and modeling and PF for Bayesian estimation in LS, the proposed approach can improve prediction accuracy, provide uncertainty representation and management for fault state and RUL, and reduce the computation for real-time applications.

Potential Applications:

This patent can be used for battery diagnosis and prognosis for lithium-ion batteries. Lithium-ion batteries are widely used in applications such as cell phones, laptops, electric vehicles, watches, etc. The global lithium-ion battery market was valued at $36.7 billion in 2019 and is projected to hit $129.3 billion by 2027, at a GAGR of 18% from 2020 to 2027. The increasing development of electric vehicles highly drives the lithium-ion battery market in recent years. Battery diagnosis and prognosis is a critical technique that can provide accurate state estimation and remaining life prediction. This patent presented a data-driven-based FDP method, which can improve prediction accuracy, provide uncertainty, representation, and management for fault state and RUL, and reduce the computation for real-time applications. This patent is especially suitable for real-time applications where FDP algorithms are deployed on portable devices or embedded systems with limited computation and storage capabilities.

Advantages and Benefits:

This patent takes advantage of DBN’s strong automatic learning capability, the uncertainty management capability of Bayesian estimation, and the cost-efficiency of the LS-based FDP framework to achieve computation efficiency and accuracy of FDP. The verification results and comparisons demonstrated that the patent significantly improved in real-time condition estimation and RUL prediction in terms of accuracy and efficiency.