Gather valuable insight from machine data for predictive maintenance and optimal process utilisation.
Digging into machine data can be overwhelming and requires a strategic and moderated approach. Sensorise’s data scientists and engineers help visualise, explore and explain sensor data in a moderated workhow format, while the domain experts can interpret the results. The collaboration creates common understanding and reduces the cost of further integrations and software development. Whenever possible and feasible the solution is developed based on engineering knowledge. Sometimes, however, there are too many variables to construct accurate physics-based models, in which case data-based models could be created by using a suitable machine learning technique.
Beautiful and meaningful visualisation
The human optic nerve has a bandwidth of 1 Mbit/sec. Dashboards and interactive visualisations of sensor readings enable a better understanding of the assets’ condition:> Time domain representation
> Frequency domain representation
> Order Analysis plots
> Correlation plots
> Statistics / KPIs, including evolution over time
Building application-specific models
When raw data is not enough: Anomaly or novelty detection solutions generate warnings for further analysis. The following algorithms have proven to be useful:> Local Outlier Factor
> LSTM (Neural) Networks
Additional data sources are often included for improving analysis results and performance of data-driven models:> Signals from SCADA system
> Information from weather stations or forecasts
> Satellite data
As these solutions are often needed in the industry, Sensorise engineers have developed internal tools which allow developing quick and cost efficient prototypes for initial technical validation of such solutions, while also providing skills to implement these solutions in production.