The Opportunity
The automotive cabling industry is governed by strict ISO and brand-specific standards, with rigorous testing required for product approval. Traditional R&D methods often involve extensive cycles of prototyping and laboratory validation, consuming weeks for fatigue and compliance tests. At the same time, differentiation depends on proprietary insulating materials and precise control over cable properties.
Slow, iterative R&D cycles
Geometric modelling, material selection, prototyping, and exhaustive compliance tests created long feedback loops. Each failed iteration meant restarting a multi-week process, limiting how many design options could realistically be explored.
Knowledge locked in people and spreadsheets
Critical know-how was embedded in individual experience and Excel-based workflows. There was no central system to capture, structure, and reuse this knowledge across projects and teams, making onboarding and standardisation difficult.
Limited trust in black-box predictions
R&D teams were used to relying on physical tests and intuition. Any AI-based approach had to earn trust by exposing uncertainty, not hiding it, and by fitting into existing ways of working rather than replacing them overnight.
The Solution
The project redefines cable R&D by combining advanced machine learning, AI-supported simulation, and optimisation in a single, scalable platform.
At its core, the platform introduces surrogate modelling: machine learning models trained on real-world data and traditional simulations to provide rapid predictions for thermal and material behaviour. Instead of relying solely on costly and time-intensive full-physics simulations, these models predict key properties within seconds, reserving detailed simulations and lab tests for the most promising candidates.
An optimisation engine explores the design space - geometry, materials, and process parameters, with objectives such as minimising cable diameter, reducing material costs, or meeting specific thermal constraints. With this setup, the system can evaluate up to thousands of design permutations quickly, narrowing the field to a shortlist suitable for physical prototyping.
A dedicated chemistry module enables virtual formulation of insulating materials. By predicting properties like hardness or humidity resistance from different polymer combinations, it reduces dependency on trial-and-error experiments and accelerates the development of proprietary compounds.
To address the legacy data landscape, database and ETL pipelines were built to extract, transform, and structure information from PDFs and Excel sheets into a central, usable format. This created a single source of truth for simulations, models, and optimisation runs.
Custom interfaces were designed to match how engineers already work, while exposing model predictions with confidence intervals and error margins. This transparency helps build trust in the outputs and supports a gradual shift from intuition-only decisions to data-supported ones, without disrupting existing R&D workflows.
The Impact
The platform does not replace physical testing; it makes it more targeted and efficient. This wider, faster exploration increases the likelihood of finding better trade-offs between performance, cost, and manufacturability, while reducing the number of physical prototypes required.
Custom interfaces give engineers a familiar way to interact with models and simulations without needing to learn entirely new tools. Multi‑objective optimisation, with explicit trade-offs between design features and cost projections, brings cost awareness earlier into the pipeline, improving efficiency before production decisions are locked in.
Beyond automotive cables, the same approach - combining simulation, surrogate models, and structured data - is immediately applicable to other industries facing similar design and optimisation challenges, such as pharmaceuticals, aerospace, or industrial components.
