This work presents a data-driven investigation into the extent to which artificial intelligence (AI) models capture physically meaningful behaviour in lithium-ion battery systems. Using operando X-ray diffraction (XRD) data in combination with electrochemical measurements, the study explores how explainability techniques can be employed to interpret model predictions and assess whether learned representations align with underlying electrochemical processes.

Rather than incorporating explicit physics-based modelling, the approach leverages AI to learn structure–property relationships directly from experimental data. By analysing feature importance and attention patterns, key regions within XRD signals that contribute most to model predictions are identified, providing insight into how AI may implicitly reflect battery dynamics. This enables a critical evaluation of the interpretability and reliability of AI-driven models in this domain.

The findings highlight both the potential and current limitations of such approaches, particularly in scenarios involving noisy data and varying states of charge. Building on this, the work outlines future directions aimed at extending the methodology from single-layer electrode analysis to more complex multi-layer cells and full battery pack systems, facilitating more comprehensive modelling of electrochemical behaviour in practical applications.

Lithium-ion batteries are the key element of electrified transportation systems during the transition to a net-zero future. However, they are inherently complex systems with electrochemical, mechanical, and electrical aspects influencing their performance and life. The manufacturing process of these batteries has up to 140 steps and needs almost 600 variables to be determined. This makes optimisation the manufacturing process and the battery performance very difficult, time and resource-intensive.

The talk will discuss the process of manufacturing from an engineering and modelling point of view and show the opportunities of AI to address some of the challenges.

Precise control of electrode thickness is essential for manufacturing high-performance batteries, as variations can affect energy density, conductivity, and overall lifespan. Inline characterisation techniques are crucial for quality control, and ultrasound-based measurements provide a non-destructive and real-time solution for thickness estimation. However, interpreting ultrasound signals is challenging due to noise, complex patterns, and material-dependent properties, making automation vital for efficiency and reliability. Our work presents a deep learning framework for electrode thickness prediction using ultrasound signals to ensure uniformity across the production line. The proposed model integrates multi-scale convolutional neural networks (CNNs) for local feature extraction with a shifted window (Swin) Transformer to capture long-range dependencies. This design eliminates the need for handcrafted features and improves adaptability to diverse signal characteristics in real-world manufacturing. To address the limited availability of labelled data, a generative adversarial network (GAN) is employed to produce synthetic ultrasound signals, effectively augmenting the training set and enhancing model generalisation. Experimental validation on data from a lab-scale electrode production line demonstrates that the proposed method achieves accurate and real-time thickness prediction. By combining CNNs, Transformers, and GAN-based data augmentation, the framework offers a robust, scalable, and non-destructive solution for inline battery manufacturing quality control.

This work presents a depth image refinement technique designed to enhance the usability of a commercial camera in underwater environments. Stereo vision-based depth cameras offer dense data that is well-suited for accurate environmental understanding. However, light attenuation in water introduces challenges such as missing regions, outliers, and noise in the captured depth images, which can degrade performance in computer vision tasks. Using the Intel RealSense D455 camera, we captured data in a controlled water tank and proposed a refinement technique leveraging the state-of-the-art Depth-Anything model. Our approach involves first capturing a depth image with the Intel RealSense camera and generating a relative depth image using the Depth-Anything model based on the recorded color image. We then apply a mapping between the Depth-Anything generated relative depth data and the RealSense depth image to produce a visually appealing and accurate depth image. Our results demonstrate that this technique enables precise depth measurement at distances of up to 1.2 meters underwater.

Leveraging advanced artificial intelligence (AI) methodologies offers the advantage of incorporating multiple expert viewpoints, thereby facilitating a more comprehensive inspection of underwater infrastructure. However, the implementation of AI techniques in subsea tasks is hindered by the lack of extensive and diverse datasets required for effective training and inference of the AI models, emphasizing the vital need for enhanced data sharing practices within the offshore sector. The sensitive textual information within underwater survey data, such as site geolocations, water depths, mission-specific details, timestamps, and third-party data, necessitate a balanced approach to data privacy. To address this, we propose the integration of cutting-edge text detection and image inpainting techniques. These methodologies enable the identification and subsequent removal of textual regions from images while preserving the quality and natural appearance of the images. Experimental results validate the efficacy of our proposed approach in simultaneously preserving the visual quality and protecting the privacy. The removal of detected textual regions from images demonstrates less distortions, underscoring the potential of this methodology for application in offshore industry settings. This study contributes to the ongoing discourse regarding data privacy in underwater surveys, offering a viable solution to balance information sharing with confidentiality concerns.