We develop innovative diagnostic algorithms, methods and software based on models of various types, transparent or black box type, e.g. neural networks, or on unsupervised data mining methods, e.g. based on feature and patterns extraction, clustering and correlation analyses for self learning novelty and fault detection.
Our process or machine virtual sensors and diagnostic software can be embedded in custom and commercial electronic hardware for real time diagnostics, networked in several ways with the plant, e.g. CAN bus or other field bus technologies. Our solutions have been already embedded in the electronics of cars, trucks, buses, industrial and naval engines and are being studied for space-crafts telemetry analysis.
Due to their generality these methods can be applied also to infield or telemetry data from oil&gas processes and machines.
The example shown in the figures is a simple yet representative solution applied to large energy generation engines, based, in this case, on a transparent model of the mass and energy balance of the engine.
For signals preprocessing and black box models definition and identification from in field data, we use our proprietary software suite GPMAS (General Purpose Mathematical Application Server).
GPMAS allows exporting the systems architecture definition and parameters to be used seamlessly by our dynamic libraries (.dll) that implement online models and that can be used in our or other diagnostic systems.