Implementation Solution

SmartFactory manages a variety of monitoring and recording devices that can be used to monitor a production line and measure the process parameters and energy usage. This data is stored in the Big Data Platform and will be used in the development of this solution. The components of the solution include aspects such as the analysis of data collected during the manufacturing process, the analysis of energy usage in the manufacturing process, the investigation of the parameter space and identification of the fixed and variable parameters or the usage of deep learning to explore the impact of the previous states/configurations on energy and quality.

For the implementation of the solution, a four-step process was defined, which started with data collection and exploration activities, followed by the initial modelling achieving a simple model to test hypothesis. Once the hypothesis was tested expanded models for product and energy were developed, ending with the final models and documentations tested and validated.

Results Obtained

The IoT infrastructure for data collection was put in place and data was aggregated at the historian.

The time series data was collected and provided by SmartFactory through a historian data collection system. A pipeline for data analysis was developed by CeADAR to streamline the process of analysis of the big data stream – this includes pre-processing functionality with data quality verifications and mechanisms to deal with data irregularities. The data was modelled to isolate the different products that were being produced and profile the average energy usage of each product. Different models to correlate the variety of features against the energy target were employed. The energy was calculated by aggregating the Steam, Gas and Electricity usage.

Two main results which align with the goal of the experiments arose from the modelling of the data:

• The first model profiled the energy usage in the production process to see how it changed over time. This is helpful in understanding how the energy cost is linked to the time that the product was processing for and is key for Danone to take decisions in how to process and optimise the energy usage. The next two graphs show an example of the energy profiles for Electricity and Steam for one of the products. The step count reflects 5 minute intervals. The standard deviation is shown by the lighter shading and the red background indicates where only one observation was present for that stepcount in the data.
• The second analysis estimate the total average energy per product taking into account the variation of energy in the different processes for each product according to different parameters. The following plot summarising the results provided where the total energy has been calculated for each product across the three energy measures of Steam, Gas and Electricity. The error bars show the standard deviation in energy measurements for that product.

SmartFactory and Donone are finding these insights very useful for decision making in the daily operations of one of the large spray dryers. They see potential to keep developing the data analytics function as part of the SmartFactory product.