The current scenario - Critical aspects
Overall performance estimates are typically based on complex simulation models driven by discrete events and by the expertise and experience of design engineers:
- The modelling of the line requires the analysis of a significant amount of data, often fragmented across different databases and difficult to reconciliate;
- The overall performance of the line is estimated on the basis of tests carried out over medium/long periods of time, only once the line has already been implemented;
- The actual overall performance is calculated retrospectively, following a long production analysis period (typically a year or more).
The use cases developed
A Simulation and a Machine Learning module has been developed, which facilitates the configuration and the execution of simulations designed to predict the overall efficiency of an automated line by integrating with Hermes Reply’s Brick platform (in particular with its Monitoring component).
Through front-end web development users can:
- Select the list and sequence of work phases that make up the line being simulated;
- Query the predictive model capable of inspecting and evaluating data sets previously collected in the field;
- View the optimal mix of machinery corresponding to the work phases specified, which will be proposed by the system;
- View the expected overall performance based on the proposed mix of machinery.
By integrating with Microsoft's Azure Machine Learning cloud services, the platform:
- Generates various proposals of optimal mixes of machinery by analysing the performance of all the machinery capable of performing the steps indicated by the designers;
- For each proposal, the solution calculates the overall performance of the line based on analysis of the individual component (or machine) performance.
The solution architecture
The Brick Machine Learning technology solution consists of a web application published on the Microsoft Azure cloud platform and integrated with a Machine Learning algorithm implemented on Azure and consumable through a web-service.
The application architecture is implemented using the following components:
- A modern HTML 5 and JavaScript web front-end, based on the AngularJS technology;
- A Machine Learning layer implemented on the Azure platform and consumable by invoking REST (Representational State Transfer) web services;
- A data persistence layer consisting of a relational MS SQL database.