Business Intelligence System For Plants

At this stage, our team examined the existing ERP platform of the client, dived deeply into the business processes, explored alternatives and figured out the best possible technological solution. We designed a model which outlines business entities in the system and their relationships, created a detailed plan for the development of each module and agreed with the client on the workflow.

According to the requirements, we recommended the technologies that perfectly suited the client's needs. Ngx-datatable was chosen to handle complex datasets in the form of spreadsheets. It provides all features to work with a table (sort, filter, add sub-rows, etc.) and is very flexible and light. D3.js was used for data visualizations and creating responsive chart components based on dynamic data. Thus, the information was presented in an easy and illustrative way that enabled users to understand it and make conclusions much faster. For Machine Learning we chose TensorFlow.js due to its high scalability and the possibility to use its capabilities on various devices if needed.

To make data analysis faster and more efficient, we implemented a Machine Learning algorithms module. It defines the patterns of changes in plants’ work parameters and assesses their effectiveness. The algorithms automatically receive the data from the ERP system and learn which ways were the most efficient in achieving the KPIs set for each plant, workshop or even machine. This allows the module to find trends and patterns that are not seen so easily but can be efficiently used in cost optimization. We chose TensorFlow.js due to its high scalability and the possibility to use its capabilities on various devices in the future if needed.

The Machine Learning module works within the manufacturer’s perimeter and doesn’t label or store data about any branch, plant, workshop, or machine. Only statistical data is analyzed by the ML algorithms. This means that this module is secure and can’t provide the potential intruder or insider with any vital information on the manufacturing powers and items of the client.

After agreement of the work breakdown structure with the client, we got down to developing the MVP.

We delivered module by module every two weeks, discussed tasks in Slack and had weekly videoconferences in Google Meets.

Each development phase concluded with the unit and manual testing, so we could detect and fix even the smallest bugs at the earliest possible stage, preventing them from turning into issues.

After agreement of the work breakdown structure with the client, we got down to developing the MVP.

We delivered module by module every two weeks, discussed tasks in Slack and had tasks tracked in Jira.

Each development phase concluded with the unit and manual testing, so we could detect and fix even the smallest bugs at the earliest possible stage, preventing them from turning into great issues.