Online Analyzer and Data Platform

Solution Context
The ore goes through various transformations before it becomes a salable product. The factory receives minerals from different destinations every day; the production cycle that follows is dependent on origin and quality. Ore from Buizrane mines, for example, has a diameter greater than 120 cm and must first pass through crushing before being poured into the reactors. The crushed ore will be attacked by the acid H2SO4 to produce the S1 solution. The production cycle continues to end in different qualities of the final product.
Digital marketing has evolved to merge with data science to create more intelligent and optimal solutions. The old approach used by the company is showing good results, feeding the sales team with strong leads every day. However, numerous processes are done in a traditional way that comes with an OPEX cost.
The old solution includes a chatbot embedded on the website, behind which there’s a team of operators engaging with every customer landing there. This manual process can cause lead loss due to late replies and involves a monthly recurring cost related to the operators. The data history of all the conversations between the operators and leads is backed up. However, no analysis was done to take advantage of this valuable information.
The company was also partnering with Sopro for lead generation and email automation, resulting in a “per contact” cost with no guaranteed results. The contact list provided by the partner has to be reviewed and cleaned by the marketing team, and then the lead assignment has to be done manually.
Salesforce is a one-stop shop for the company’s sales activity. An upgrade to the digital marketing strategy was required in order to fill these gaps.


The solution consists of two distinct parts:
Mechatronics part: These are two solenoid valves and pumps connected to the liquid treatment chamber which will make it
possible, periodically, to change the sample in the online analyzer
tank.
● Artificial vision part: The sampling tank is equipped with ahigh-definition camera with a white LED lighting device to capture the color of the liquid in the tank. The principle is to support the measurements of the online analyzer with a supervised learning model to control the process.
All the components of the online analyzer tank have thermal isolation work and an analysis of the reactivity of the materials was carried out to ensure that the corrosivity of the liquid medium does not deteriorate the components and that this does not lead to the leaving of additional residues from the sensors.
The realization of the solution was carried out according to the following steps:
● Setting up a calibration environment: This environment consists of a tank that has all the analyzer’s components online. This tank will be in a controlled environment and will allow us to detect the values observed under nominal conditions and key stages of the liquid.
● Installation of the camera and constitution of training data: We will add an HD camera to the tank to capture, in a controlled environment, the various images corresponding to the state of the liquid. This will allow us to train learning models to detect these key milestones.
● Installation of the necessary thermal insulation and assembly of the online analyzer: At the end of this step, we have a functional analysis unit with solenoid valves and pumps connected to the enclosure.
Data Platform
The project consists of two main phases—real-time monitoring and historical data processing. Each of these phases is divided into several sub-phases.
Phase 1: Real-Time reporting
The main phases that constitute real-time monitoring are as follows:
● File recovery: This phase consists of constantly listening to the target directories to recover files when they are created.
● File Processing: Retrieve key columns from each file and send a snapshot of retrieved values to Power BI.
● Reporting: This phase consists of creating the visuals requested by the group and associating with them the values received in real time from the files.
● Tests and deployment: Continuous testing of the correct functioning of the solution and deploying the final solution within the group.
Realtime report

Phase 2: Decisional Project:
Modeling: The data sources and their meaning being assimilated; we designed the entire data warehouse. The model must align perfectly both with the expectation of the group and with the feed of the flows through the ETL.
● Creation of the OLAP cube: As the cube is a priority for the group, we have chosen to ensure its deployment before feeding the Data Warehouse with real data. To do this, we generated data to test the proper functioning of the OLAP cube.
● Feeding ETL flows: During this phase, we created packages for transforming and loading customer data into the DDS. We also managed the quality of data from flat files (Data Quality Management) before storing them.
● Reporting: We fed the dashboards already created directly through the OLAP cube.
● Tests and deployment: Continuous test of the correct functioning of the decision-making chain at the customer’s premises.
