Hy-Met
Hy-Met
Hy-Met is a diagnostic sensor company, developing state-of-the-art measurement technologies to meet the current and future challenges faced in electric vehicle manufacturing, hydrogen infrastructure, and other emerging energy sectors.
Demand for Lithium-ion batteries, driven by growth in electric vehicle and energy storage markets, is expected to increase six-fold globally in the next five years. Battery manufacturing, however, remains fraught with challenges in expensive end-of-line inspection processes; typical batch sampling leads to scrap rates as high as 30% and wasting of valuable energy and rare earth materials.
Hy-Met has addressed this with a rapid, real-time battery inspection solution, “HyLite”; employing non-contact ultrasound, which is cheaper, faster, and more sensitive to certain manufacturing defects than the commonly used end-of-line tests.
The Problem
HyLite aims to employ sophisticated data science and machine learning techniques to identify defects and characterise battery cells. Hy-Met’s problem lies in the lack of available data in the battery manufacturing sector required to trained supervised machine learning algorithms. Most defective batteries never leave the factory as manufacturers are hesitant to release defective batteries (even for R&D purposes).
Therefore Hy-Met seeks assistance in developing mathematical models to simulate the ultrasound signal response from a range of healthy and defective cells; representing various cell geometries, chemical compositions, and defects. This would enable Hy-Met to generate a sufficient volume of synthetic and augmented data with which to train machine learning algorithms to detect and classify defects.
Thus far, Hy-Met has demonstrated the ability to qualitatively discriminate battery samples via differences in their response to pulsed ultrasound, as demonstrated in figures 1 and 2.
Key areas of focus for addressing this problem lies in the modelling of ultrasonic wave propagation, including boundary conditions and interfaces, and accounting for the material properties of the batteries. Starting points and considerations could include:
- Modelling wave propagation through lattice structures, considering periodic and aperiodic arrangements and the material anisotropy and heterogeneity.
- Studying reflection, refraction and transmission of acoustic waves at the material interfaces as well as their attenuation and dispersion as they travel through the battery.
- Incorporating elastic properties of the battery materials, considering inhomogeneities such as voids and varying density and relating these to the presence of defects, battery health or the state of charge.
Solutions for the implementation of numerical methods for solving the proposed models and the design of experiments to validate the theoretical models and numerical simulations will also be required. Hy-Met can provide ultrasound signal response data for the labelled / well-understood cell samples currently in its possession for comparison with simulations. Due to Hy-Met’s proximity to Birmingham (Tyseley Energy Park), it may be feasible to run experiments during ESGI to generate bespoke data and test hypotheses in real time as the week progresses.
Outcomes from this problem could include enhanced models for understanding wave propagation in lattice structures, with the desired extension of this being the development of predictive tools that can accurately determine battery health and internal properties based on acoustic wave analysis. Guidelines for interpreting acoustic wave data for battery diagnostics are an additional transferrable outcome.
Challenges may present in the mathematical complexity of wave equations in anisotropic and heterogeneous media, in the computation resources required for numerical methods or in accurately correlating the experimental data with the proposed theoretical model outputs.
Figure 1: Ultrasound signal response for defective (blue) and non-defective (orange) cells, using contact ultrasound (left) and non-contact (right). It can be seen that the signal decays more quickly in the tail for defective cells.
Figure 2: “C-scan” map of the dominant frequency in the Fourier domain for three different battery cells: pristine (left), partially gassed (centre), fully gassed (right).