EARLY FAULT DETECTION OF A GAS TURBINE FOR ENERGY PRODUCTION
Gas turbines are very critical and expensive components of the energy production industry. The unavailability of a gas turbine can result in huge economic losses, in the order of millions of euros per day: it is therefore fundamental to avoid turbine failures and to reduce as much as possible the turbine unavailability due to maintenance actions by intervening at the optimal time.
To reach this goal, Aramis has developed an ensemble approach which, by exploiting statistical and machine learning methods, allows accurately online monitoring the turbine behavior for identifying the onset of the turbine degradation in advance with respect to traditional approaches based on fixed thresholds.
The developed fault detection approach provides the early detection of the turbine degradation onset. This information allows a safer management of the turbine, avoiding critical failures and reducing the maintenance intervention time and costs.
PART-FLOW OPTIMIZATION FOR A MAINTENANCE SERVICE CONTRACT OF GAS TURBINES
In a maintenance service contract of gas turbines, capital parts are periodically removed from the gas turbines and replaced by parts of the same type. This entails that the flow of every turbine’s part needs to be optimized for reducing the maintenance costs and maximizing the revenues.
Aramis has developed a reinforcement learning algorithm for optimizing the sequence of repair and replacement actions of capital parts. The developed approach, by considering the remaining time up to the end of the contract, the availability of spares and the costs related to repair actions and failures, leads to the minimization of the costs of the maintenance service.
The developed reinforcement learning approach allows identifying a near-optimal sequence of actions is found in a feasible time avoiding the combinatorial explosion of possible cases.
A SENSIBILITY ANALYSIS BASED FRAMEWORK FOR UNSUPERVISED FAULT DIAGNOSTICS OF GAS TURBINES
Many companies rely on anomaly detection algorithms to detect Gas Turbines anomalous behaviors, which, however, do not identify the component or system that is responsible for the malfunction. This requires a large effort by experts for the identification of the causes of the anomaly, especially for complex systems with hundreds of signals from large fleets of gas turbines installed all over the world.
To overcome this issue, Aramis has developed several sensitivity-based approaches, which allow estimating the impact of each signal to the detected anomaly and, thus, identifying the group of signals responsible for it. The procedure can be easily integrated with already-existing fault detection tools.
The developed fault isolation sensitivity-based approach has proven effective in identifying the signals responsible for the anomaly in several case studies, in case of both short and long time-scales.