AI-empowered Supply chain mapping, prediction and optimization for sustainability

This task involves developing and implementing GNN models to map the entire supply chain network for audio equipment products. The GNN models will represent suppliers, manufacturers, and components as nodes, with their interactions as edges in a graph. First, we collect from various sources, extract meaningful features and build the graph to represent mode – edge characteristics. Next, we develop GNN model to learn diverse, implicit relationships or dependencies between suppliers and manufacturers, regardless they are close by or multiple hops away from each other. We then leverage this learnt knowledge to identify, for a given new product component, all relevant down- and upstream suppliers, which enhances the visibility of the entire supply chain.

This task utilizes the full-fledged graph built in Task 3.1 and leverages predictive AI techniques (regression models, temporal models) to forecast future events and trends within the supply chain, such as demand forecasting, supplier risk assessment, and inventory optimization. Our activities include analysing historical sale data, market trends, and seasonality for forecasting demand, evaluating supplier risks (e.g., supply shortage), and/or predicting optimal inventory levels.

In this task, optimization techniques (such as Genetic Algorithm), will be employed to provide actionable recommendations for optimizing the supply chain based on predictive insights from Task 3.2 and identified OEMs from Task 3.1. The proposal includes supplier selection and route optimization, which on one hand satisfy all requirements in a given BOM, but on the other hand offers benefit on transport cost and delivery time.

This task focuses on detecting and visualizing common issues in audio products using Natural Language Processing (NLP) and machine learning (ML) algorithms. First, we develop a web scraping tool to collect data from online sources, cleaning and pre-processing the data for subsequent in-depth analysis. After that, we harness ML algorithms like Latent Dirichlet Allocation (LDA) for topic modelling, which reveals technical or quality issues corresponding to different severity levels, based on users feedback.

In this task, build visualization tools to visualize the observations or learnt insights via data analysis, predictive models, or optimization initiatives, obtained in all tasks, from Task 3.1 to Task 3.4. More specifically, we create dashboards and heatmaps, which intuitively plot end-to-end supply chain for a given component or product, highlight nodes with bottlenecks (e.g., high risk of shortage or overload), visualize optimal logistic routes, showcase top frequent/critical technical issues, or display market trend (capacity/demand) in the near future. These tools facilitate manufacturers and their staff’s understanding of the entire supply chain’s behaviour, therefore assisting them in planning and decision making.