DCs & Workpackages
DC1 will focus in the development of more sensitive, accurate, robust and interpretable gear diagnostic indicators, based on signal processing and AI techniques, for monitoring gear health, estimated via (indirect) measurements, such as accelerometers, acoustic emissions, microphones and encoders. The algorithms will link, qualitatively and quantitatively, the direct and the indirect measurements. The availability of real degradation information will allow AI algorithms to find better correlations between the features computed on indirect diagnostic indicators and a better thresholds setting. Improved features and better threshold setting will allow for an earlier detection of faults (increased sensitivity) and will lead to the optimization of the ratio between the false alarms and missed detections (increased accuracy) and to a higher trust of the monitoring systems. It will also allow to correlate the defect size to features extracted from indirect measurements resulting in improved interpretability. Additionally DC1 will work towards the developing of an automated vision system for capturing of surface defects (micro and macro pitting, cracks, scuffing, spalling, etc.) initiation and for monitoring gear defect evolution. The vision system (HW/SW) will be designed to deal with both splash and spray lubrication pertaining to different lubrication regimes and also in starvation. The vision system will be able to continuously take images of all gear flanks with a frequency of at least 1 flank per second. Most industrially relevant gear types are spur or mainly helical gears and, for this reason, these gear types will be targeted. The image will be processed automatically using AI algorithms that have been trained to recognize defects on metallic surfaces. These direct observations of fault initiation and evolution will result in a dataset that will constitute the ground truth that can be used for labelling of indirect measurements. The methodologies will be tested, evaluated and compared with state of the art techniques using dedicated test rigs of planetary gearboxes.
Innovative aspects: Novel diagnostic indicators validated based on ground truth measurements, development of a vision monitoring system for drivetrains.
DC2 will work on the development of a physics inspired machine learning technique trying to combine the advantages of purely data driven approaches with the advantages of model based approaches. Advanced drivetrain models, consisted of submodels of bearings, gears and shafts, will be developed including faults such as pitting and spalling. Multibody and FEM models will be combined in order to achieve the needed accuracy. The developed models will be merged with data driven approaches in different directions. As usually there are not enough data in order to train a machine learning approach at the faulty classes, the validated model will be used in order to generate artificial data which will cover the measurement space and can be used for the training of a machine learning methodology. Additionally the embedding of the physical model in an artificial neural network will be considered, either by preparing part of the architecture of the network based on the physical model, or by putting the physical model and the data model sequentially, so the data model can practically model the missing physics of the physical model. The combination of the two models will lead to a digital twin which can be further used for diagnostics and prognostics. The models and the methodologies will be tested, evaluated and validated using dedicated test rigs of drivetrains of different power.
Innovative aspects: Physical inspired machine learning approach for condition monitoring, Hybrid models.
DC3 will work on the design of novel HIs, considering explicitly the physics of degradation. Health indicators (HI) are fundamental quantities at the basis of current diagnosis and prognosis methodologies. Despite remarkable progresses in health monitoring boosted by new technologies (IoT, new sensors) and AI, most approaches still rely on the use of rudimentary HIs defined more than half a century ago, when the main motivation was to provide metrics that could be easily calculated with the low computational resources of that time. Many popular HIs have traded simplicity against physical relevance and, as a consequence, it turns out difficult to tailor them to monitor specific degradation processes. Rather paradoxically, HIs with limited informational content are used as the inputs of extremely sophisticated machine learning algorithms (such as regressors and classifiers), yet constituting the weakest link of the chain. DC3 will work towards the construction of a mathematical mapping from physical multidimensional quantities such as surface topology and local mechanical properties to a scalar metric that can be calculated from the measurement of the dynamic behaviour of the structure. Models of tribology will be used to correlate the dynamic response of a structure to local properties of damaged surfaces of contact (gears, rolling element bearings). Fracture dynamics and fatigue models will be considered to construct metrics of damage. DC3 will define statistical indicators, as functionals of vibration signals, that maximally reflect the physical information. The methodologies will be tested, evaluated and validated experimentally using a testbench for fatigue analysis of rolling element bearings and tribometers.
Innovative aspects: Correlation between tribometer data and vibration measurements
DC4 will work on the development of a Lubrication Quality (LQ) parameter to be used to determine if there is a risk of wear or fatigue damage in rolling-sliding contacts such as those found in bearings, gears, cams, and other non-conformally shaped machine components. Today, the film parameter, the ratio between estimated film thickness and surface roughness height is used as a measure to determine the lubrication regime the machine component operates in. The regime is a direct measure of LQ. This film parameter has been found to be a very poor measure of the regime but it is still used extensively because it is very easy to be applied. Recently we developed a new film parameter which is much more accurate in determining the transition between high and low LQ. This film parameter is more complex to determine and requires advanced surface topography analyses to be estimated. The first tests show, however, an excellent capacity to determine when lubrication quality changes. Still, the new parameter is in an early stage of development before it can be an industrial standard. DC4 will investigate further how the new film parameter can be used to determine the risk of wear, micro-pitting, pitting, or scuffing. By using advanced condition monitoring based on electric contact resistance, optical load sensing, and acoustic emissions the DC will study the failure of non-conformal elastohydrodynamically lubricated contacts, looking towards how the new film parameter can be used for prognostics and diagnostics of failures. A new gear test device will be developed for this purpose. Machine components lubricated with fossil-free water-based lubricants will be also considered.
Innovative aspects: Investigation and further development of a new film parameter to determine the risk of damage, experimental investigation of non-conformal elastohydrodynamically lubricated contacts.
DC5 will work on an innovative approach, which is clearly required to reduce full-film friction from current viscosity-based solutions. At the same time, wear of the components should not be increased. One way to significantly reduce full-film friction is to employ various surface modification technologies, from nano to macro-scale, which can reduce full-film friction through a solid-liquid interface mechanisms and can gain substantial friction reduction. Recent studies indicate several possibilities to modify the contact interface that need to be explored. However, how these interface mechanisms affect the film formation, shear, thickness, load distribution, and consequently the friction and wear is currently not known; this knowledge is needed to integrate the interface boundary films into the modelling of contacts in mechanical components in order to ensure the machinery efficiency and reliability.
Innovative aspects: Determining boundary films key interface properties to reduce full-film friction, introducing boundary layer effect into contact modelling.
DC6 will focus on the analysis of the failure behaviour of rotating machinery assets during operation under uncertain conditions, targeting to the estimation of the RUL. The DC will cover the different technological aspects across the full value chain of RUL estimation and structural reliability condition monitoring. First, an analysis of the failure mode, including their relation to manufacturing processes and material defects and their criticality will be conducted, with the aim of deciding which metrics would be necessary to determine the degradation stage of the asset. Different types of failure modes from material defects, their identification from sensors measuring different physical quantities, and signals will be evaluated and the ones that provide a higher value of observability on the prognostics problem will be selected. The DC will use state-of-the-art techniques to determine the degradation from virtual models and signal processing to focus on the development of data fusion techniques. Afterwards, a methodology for trend analysis and RUL estimation will be defined, initially based on some a priori hypothesis (such as stationary operating conditions). The methodology will be further extended to cover changing operational contexts, executed maintenance actions and the influence of other systems. The algorithms will be tested, evaluated, validated and compared on data gathered from a lab-scale test bench as well as on industrial signals.
Innovative aspects: Multiphysics/multisensor prognostics, Prognostics under uncertain conditions.
DC7 will work on the health monitoring of aircraft engine planetary gearboxes which is an emerging research topic, attracting growing interest from aerospace companies. Planetary gears, ubiquitous in mechanical systems, convert speed and torque in a power transmission chain, of which they often constitute a weak link in terms of wear and tear. DC7 will explore the research approach of breaking down a vibration signal, measured on the structure supporting the planetary gears, into a sequence of elementary contributions related to the different moving parts of the gearbox and combined in a more or less elaborated way to end in the final measured signal. Usual ways of achieving this task rely on the hypothesis that the measured signal results from multi-modulations of a main frequency called meshing frequency. The decomposition operation mentioned above then boils down to demodulations but the classical methods such as Hilbert Transform are inappropriate because the multi-modulations generate aliasing frequencies. DC7 will revisit the complicated problem of mechanical signal decomposition, leveraging upon recent advances of machine learning and signal. DC7 will investigate the use of auto-encoders as a way of building a low-dimensional representation of the signal, which can be expected to contain the useful physical information. The obtained latent space will then have to be interpreted and understood, which makes the proposed research topic especially difficult. Based on this latent space the DC will propose the construction of health indicators. These indicators will be obtaining from signal processing tools.
Innovative aspects: Condition monitoring of aircraft engine gearboxes based on Autoencoders and latent variables.
DC8 will work on innovative techniques for the optimal management of agricultural vehicles in an open field based on the determination of the machine condition while in operation and by considering sudden changes of weather conditions. Following agriculture 4.0, nowadays agricultural vehicles are very sophisticated and equipped with many sensors and actuators. Making the optimal decision is usually very difficult because information is not easily obtained and integrated. Information about the technical condition or health status of the machine, the cost of maintenance or loss of production, and customer information, are not defined in the same units and are not provided on a consistent time scale. Some data is constantly updated e.g., health status data, but data like customer information is usually extracted from a historical data that is fixed over time. DC8 will develop an intelligent maintenance system, based on intelligent data processing systems, in order to exploit heterogeneous data. DC8 will develop methodologies for selection of minimum set of sensors, optimal data collection and optimal data fusion for multiple sensor detection systems.
Innovative aspects: Multi-time scale/multisensor AI based maintenance strategy for health monitoring under varying operating and environmental conditions.
DC9 will work on the development of AI methodologies for diagnostics and prognostics of industrial gearmotor drive units. Diagnostics and prognostics state of the art technologies have two key limits for a wider diffusion in all industrial sectors: a) the cost to customize the solution for each application and b) the reliability of the RUL estimation of the product when a potential failure has been identified. DC9 will try to answer those two challenges by 1) developing a customized low cost multisensor for data acquisition and a machine learning approach to reduce commissioning to set up the condition monitoring device for each specific applications, 2) exploring heterogeneous signals with novel ML/AI based techniques to detect anomalies or new potential faults, 3) developing a new approach for RUL calculation based on AI techniques to improve the reliability of the forecasting versus traditional digital twin approach using physical algorithms, 4) estimating through AI techniques of the impact of the end user application duty cycle on the laboratory test for product characterization to improve product design and calculations.
Innovative aspects: Estimation of RUL through AI/ML techniques utilizing low cost multisensors custom devices, Correlation between application duty cycle and factory test/calculations.
DC10 will work on gear contact modelling that will enable the capturing of the gear system dynamic behaviour under the existence of various gear degradation types and levels of severity. Faults of different sizes and at multiple locations at the tooth root and the flank surface will be considered. A numerical model to compute a time-varying mesh stiffness will be developed. The analytical gear stiffness model will account for the helical gears with tooth flank modification by considering slicing and tooth coupling effects, where several defects are exhibited over the multiple teeth. Manufacturing errors such as the form deviation at an individual tooth will be considered as well as the gear misalignments from the system deflections at a multibody simulation environment. From the vibration signals captured by this simulation tool, the changes in the system dynamic response under gear degradation will be investigated. Advanced data processing techniques will be developed to filter out the noise effectively to capture the residual signals. DC 10 will develop useful features to represent the gear system's health status. The developed mesh stiffness model will be tested, validated and evaluated by conducting gear pair dynamic tests on a specially designed gear precision back-to-back test rig.
Innovative aspects: Gear time-varying mesh stiffness accounting for gear degradation, Feature extraction to represent the gear health condition from system-level multibody simulations.