Maintenance of engineering assets is often done via reactive maintenance where ad hoc repairs are carried out when necessary. Condition-based monitoring is the baseline of preventive maintenance (PM), as opposed to reactive maintenance. Indeed, maintenance should be seen as a strategic business function as opposed to a necessary evil. Predictive maintenance goes one step further by leveraging on AI/ML models, which interpret asset condition data, to prevent impending failure and help schedule maintenance only when it is needed.
An aircraft or spaceship contains many different types of assets and it is impossible to develop a one size fit all PM solution. Versatility is difficult as every use-case is different. Collins Aerospace’s survey on PM over the past 25 years showed lackluster results. The problem is big data; many companies implementing PM ends up with tons of records and don’t know what to do with them. Damage prediction is often difficult to extract from those data and requires engineers’ expertise, a service often charged to operators at a high price by OEMs. Last, instrumentation often poses problem as operators are not usually willing to install sensors on their assets as it often requires integrity testing and validation by the authorities.
Big Data to Smart Data
We are aware that, given the current state of the art, Predictive Maintenance cannot address every possible use-case scenario. So we chose Structural Health Monitoring and rotating assemblies because they represent the assets with the biggest impact on operator’s budget. We help customers transition from big data to smart data. AI is not only data-expensive, it is data-format dependent: when each dataset is carefully defined we call it “labeled”. The problem with big dataset is that most of it is either wrongly defined or “unlabeled”. At Vega MX, we put a lot of effort in database preprocessing and adaptive AI such as Semi Supervised Neural Networks. We also combine, whenever possible, AI with scientific models in order to cater for the lack of usable data. This is true with the model we developed with our partner for composite material for instance. To facilitate aircraft instrumentation validation, we teamed up with aerospace key players.
Physics-Informed Machine Learning
With the vision of deploying a predictive machine-learning based tool for structural health monitoring, we developed a physics informed data-driven adaptive multi-fidelity approach for modeling crack initiation and propagation in structural aerospace components. We propose to use peridynamics, a nonlocal reformulation of classical continuum mechanics suitable for simulation of crack initiation and propagation in materials, to predict failures on composite materials under extreme environment. However, due to the significant computational cost of peridynamics compared to classical models, it is not feasible to run large-scale peridynamics simulations in real time. Therefore, we propose to employ high-fidelity peridynamics modeling representations only in restricted critical regions, while representing other regions with less computationally demanding, lower-fidelity classical models. A data-driven model adaptation strategy based on sensor and experimental data is employed to determine the critical regions as well as their evolution in time.
Easy and intuitive estimation of Remaining Useful Life (RUL)
In every airliner, all current cautions and warnings alert the crew of a technical event at a time t, but there is no prediction of impending failure of damage propagation. This is why, for in space NDE, the symbology and warning philosophy must take into account prediction and prognostic. Another important point to consider is maintenance. On aircraft, master cautions and warnings solved during a flight are linked to a dispatch message. This message is intended to be read by the maintenance engineers upon arrival. But during deep space exploration, astronauts must perform maintenance and inspection onboard with limited ground support.
Want to know more ?
The Oil & Gas, wind and power generation industry has a rather narrow set of assets and predictive maintenance has proven to be a great success. Results speak for themselves: USD 9 per horse power (as compared to 18 with reactive maintenance), 70% elimination of breakdown and 35% reduction of downtime (Rolland Berger Nov 2014).
The Global Predictive Maintenance Market size was valued USD 4.2 billion in 2021 and is estimated to reach USD 15.9 billion by 2026 (Markets & Markets). Artificial Intelligence, Machine Learning and cloud computing are the major enabling technologies supporting a strong growth rate. Although Civil Engineering and factory monitoring will take the biggest share of the market, Aerospace companies will also benefit from it in terms of maintenance cost savings.
Regarding Structural Health Monitoring, a key technology enabling predictive maintenance, Grand View Research Inc. estimates a global SHM market of USD 4.34 billion by 2025 with the biggest share taken by civil infrastructure; Aerospace, energy and mining coming next.
Legacy aircraft used beyond service life in major armed forces also constitute an important part of the aerospace market.
Last, the development of OSAM, ISAM (On-orbit Servicing Assembly and Manufacturing, In-space Servicing Assembly and Manufacturing), tourism and life sciences will depend on a fleet of dependable spaceships capable of carrying out, without failure, cargo and crew return flights at a high frequency and space stations with a much higher level of autonomy. The market is at its infancy but is sure to develop fast under the impulse from major Space Agencies.