top of page
Image by SJ Objio
Title: Advanced Analytics and AI Application to Predict Concentrate Quality
Organization

Vale - IBM

Period

July 2022 - July 2024

Status

Completed

Est. Budget

USD 1.6M

Est. Business Benefit

USD 10.0M

Indicative Value-to-cost

6.2x

Keywords

Concentrate quality, AI forecasting, IPQA, mass recovery, laboratory testing optimisation, process analytics

Benefit basis

Estimated value from reduced laboratory testing, reduced quality deviation penalties, improved mass-recovery forecasting, and faster service-order decisions.

Business Challenges 
​The in-process quality assurance team needed to predict concentrate quality and mass recovery before final service orders were issued to third-party concentration plants. Laboratory testing was expensive and time-consuming, while historical data was incomplete or unavailable for some material blends and operating conditions. The growing number of concentration service orders increased the need for faster, more reliable quality prediction and simulation capability.
Expanded Technical Solution
The project delivered an AI-based concentrate-quality prediction framework that analysed more than 20 process, material, operational, and historical variables. The model estimated expected concentrate quality, mass recovery, and uncertainty before material arrived at the plant. The solution combined process knowledge with machine-learning methods so that the IPQA team could simulate product quality, reduce unnecessary testing, guide third-party plants, and provide better information to planning and commercial teams.
Technical Work Packages/Methods
  • Integrated production history, feed characteristics, ore source, blend information, plant operating parameters, laboratory assays, mass recovery data, service orders, and third-party plant performance data.
  • Applied data-quality profiling, missing-value treatment, feature selection, and physics/process-informed constraints.
  • Developed predictive models using methods such as Random Forest, gradient boosting, regularised regression, neural networks, and ensemble blending.
  • Estimated confidence bands for quality predictions and flagged cases requiring laboratory confirmation.
  • Created decision-support outputs for service-order preparation, third-party plant guidance, and quality-risk management.
Key Deliverables
  • Concentrate-quality prediction model using more than 20 effective parameters.
  • Integrated data pipeline for IPQA, production, laboratory, planning, and service-order data.
  • Dashboard/reporting layer for predicted quality, mass recovery, and model confidence.
  • Operational guidance for laboratory-test reduction, quality-risk review, and model retraining.
Results and Value Created
The project reduced dependence on slow and expensive laboratory testing for routine quality prediction while improving the quality of information available for service orders and third-party plant management. It also supported earlier quality-risk identification and better alignment between operations, IPQA, planning, and commercial teams. The estimated benefit reflects reduced testing cost, reduced quality penalties, improved recovery planning, and faster service-order preparation.
bottom of page