Machine Learning for Control Improvement

WHAT WE DO

ISC is at the forefront of researching and developing effective solutions to enhance controller capabilities by integrating cutting-edge Artificial Intelligence (AI) technology, in particular data-driven Machine Learning (ML) tools, and advanced control theory paradigm.

Data-Driven AI: Machine Learning

Data-driven AI techniques, particularly machine learning (ML), are increasingly sought after for their ability to automate the management of diverse data sources and uncover hidden correlations within the data. These capabilities empower technology adopters to extract valuable insights from the vast amounts of data they accumulate. Machine learning techniques are powerful because they can automatically learn from data without requiring any prior knowledge of the underlying physical system.

data-driven ML development

Data-Driven Machine Learning

Techniques

  • Meta-heuristic optimization - These techniques provide a nonlinear optimization framework permitting to solve complex optimization problems by limiting the magnitude of the sub-optimality provided due to effects of local minima.
  • Fuzzy Logic - It is a method to identify the unknown parameters of a highly complex and nonlinear system, such as a EV battery. It does not require a mathematical model but only uses the input data and the fuzzy rule base.
  • Neural Networks (NN) - Inspired by the human brain, NN is a framework of many different machine learning algorithms to perform different tasks. The NN has self-adaptability and learning abilities to establish a highly complicated and non-linear system, such as a battery.
  • Support Vector Machine (SVM) - SVM techniques have attracted considerable attention in recent years and have become a powerful tool for solving regression problems in nonlinear systems through different kernel functions and regression algorithms to transmute a nonlinear model into a linear model.

AI for Control

Because of its flexibility, modern ML based AI solutions can be used to overcome the limitations of modern control theory, enhancing the performance of control algorithms across different fields, including Robotic Process Automation, Decision Management, Adaptive Manufacturing.

Control-aimed ML solutions can be classified in two main categories, off-line and on-line ML-based control solutions:

  • Off-line solution - Typical solutions consider AI-based automatic calibration systems, data-generators for Rapid Prototyping (RP) test and algorithms for off-line estimation of unmeasurable signals/parameters.
  • On-line solution - the set of applications involving the use of AI algorithm for working in conjunction with controllers/observers developed according to standard control theory for defining a closed loop AI-based control system.

EXAMPLES OF OUR WORK

Data-Driven Driver Intent Prediction

ISC has been involved in a project to improve the performance of hybrid powertrains by exploiting information from on-board sensors and communication systems, combining with external systems like GPS and weather forecasting systems. This allows an improvement in the powertrain performance to be achieved by changing the controller design philosophy to optimize range and reduce component degradation in for example batteries and fuel cells. ISC has developed an algorithm based on Machine Learning and Data-Driven techniques to convert data from different systems to use in the hybrid powertrain control and monitoring system. Support Vector Machine modelling and identification methods were developed to adapt the controller in uncertain operating conditions to overcome limits and improve the performance over the more traditional baseline controllers.

Digital Twin for Electric Vehicle Monitoring

ISC has applied NN-based algorithms for solving different identification and estimation problems, by developing ad-hoc networks able to replicate nonlinear dynamics by a pure data-driven approach according to the Digital Twin paradigm. The solutions reduced the size of the network, and related number of parameters and the gains to calibrate, reducing the training time for setting up the network.

Data-driven Estimation for Battery Systems

This Scottish Enterprise funded project focuses on optimising the performance and control of batteries in various transport applications from heavy goods vehicles to marine systems. The work exploits recent advances in AI to improve the quality of the estimators needed for determining critical variables of the battery, such as state-of-charge and state-of-health. The new data-driven, AI/ML-based estimation algorithm improves the estimation performance by exploiting advanced statistical theory and data-driven techniques. The resulting design tool reduces the development effort and cost.

WHAT OUR CUSTOMERS SAID ABOUT OUR WORK