Control Algorithms
Design for Innovative Products
We specialize in the design, implementation, validation, and maintenance of control algorithms across the entire lifespan of a product. Our expertise extends to system-level specifications, developed collaboratively with our clients, as control algorithms frequently have a defining impact on the product as a whole.
We have extensive experience in all the relevant control algorithm families, including:
- SISO and MIMO closed loops
- State machines based on standard discrete and continuous automata models, tailored as needed for specific project requirements
- Model predictive, robust, optimum, and various other model-based control theories from literature
Addressing the estimation aspect of control challenges, we confidently apply several digital filtering techniques. Our expertise ranges from ARMA structures for fundamental SISO requirements to advanced model-based, optimal observers such as Kalman filters and their nonlinear extensions.
We leverage AI-based tools to:
- Bridge modelling gaps where increasing the complexity of white-box models is not economically feasible, but a black-box approach proves effective with sufficient data quantity and quality
- Offer a secondary qualifying opinion on specific parameters estimated by other algorithms, enhancing overall algorithmic robustness and reliability
Case studies
Goal statement:
In a thermal energy storage plant, control algorithms play a primary role in achieving:
- Best Round Trip Efficiency (RTE)
- Compliance with grid norms
- Plant operational capabilities
At the same time, such algorithms also keep plant operating parameters within the design limitations of plant components, such as turbomachinery, heat exchangers, tanks, piping, etc.
Solution provided:
A comprehensive dynamic model of the plant is built, integrating equations from different domains (thermodynamic, electrical, etc.). The optimum trade-off in model complexity is set taking into account PFDs and P&IDs. Such model is built to be interfaced in closed loop with the control software. Hence a framework for developing, maintaining and versioning the model, the control software and its tunings is put in place. Control software architecture and functionalities are then developed with fast iterations allowed by the model-in-the-loop simulations. The simulation iterations are made more powerful by a pre/post processing toolchain. The preprocessing allows agile parallel execution of a high number of simulations, which combined represent the whole spectrum of the plant operations and environmental variabilities, without relying on human input unless strictly needed. The post-processing toolchain operates on the executed simulations to produce automatically generated reports, optimizing the use of analysts’ time.
Control functionalities integrate the whole spectrum of traditional closed loop solutions, model based optimum controls, estimation of non measurable variables, along with the discrete domain solutions such as finite state machines and algebraic logics. Algorithms are tuned to comply with requirements from all the above-mentioned sources (RTE, grid norms, operability). Compliance to requirements is documented for all internal customers and external partners.
The control code is documented for EPC specifications where a DCS target is contemplated, while other code fractions are used to directly generate and deploy a PLC code. Field commissioning and performance
validation follow.
Problem statement:
The state of charge of a battery is a non-measurable internal state. Measurable variables of a battery include voltage, current, temperature. Mathematically, the state of charge is the integral of the past current, both charging and discharging. However computationally a simple integration is not feasible because of the intrinsic instability of the integration process (even a slightest bias would result in a diverging error).
The state of charge is also correlated with the battery voltage. The strength of such correlation depends on the specific chemical makeup of the battery, but generally the voltage is much more affected by present and short-term past current than by long-term state of charge variations.
Solution provided:
Through lab-controlled tests, where the state of charge can be considered known, the battery hysteresis dynamics, i.e. the battery voltage response, is modelled as a dynamic function of current, temperature and state of charge. For the real time estimation on vehicles two models are then used. An inverted model returning state of charge as a function of current, voltage and temperature is used for short-term open-loop estimation.
A direct model returning voltage as a function of current, temperature and state of charge is used to compare the estimated voltage with the measured voltage. The voltage estimation error thus obtained is used for long-term, closed-loop, slow correction of the state of charge estimation. Combining the direct with the inverse model leads to a valid real-time state of charge estimation, allowing for optimum use of the hybrid power train. The estimation algorithm is validated in lab conditions and on vehicle.
