can lose prediction accuracy over time, whereas the machine learning algorithms built into OPTIX constantly tune predictive models and preserve prediction accuracy in real time. Each OPTIX model is supported with a self- diagnostic prediction accuracy monitor, which calculates the difference between the prediction and actual value of the quality measurement. Upon successful implementation of the predictive model, deep learning and advanced data analytic methods are used to develop a prescriptive recommendation engine. OPTIX prescribes real-time actionable insights for optimising the manufacturing process. Operators can use data- driven recommendations to maintain manufacturing conditions or target optimised machine efficiency. SIGNIFICANT RETURN ON INVESTMENT Implementation of OPTIX provides solutions that cannot be obtained any other way and enables value-creating opportunities not previously available. These opportunities are illustrated in Figure 5 and detailed on the right. Reduced error. OPTIX real-time predictions have inherently less error than lab measurements. When the prediction is used in place of the lab measurements to make quality control decisions, the system variability decreases because the testing error is no longer transferred into the process. Reduced variability. Using the real- time predictions generated by OPTIX, operators can make optimal adjustments to key process variables to positively influence the key quality parameter. By avoiding typical reactions to overfeed basis weight and/or chemistry or to reduce speed, overall variability is reduced. Improved quality consistency. Real-time monitoring of product quality allows for informed decision-making. Reduced process variability and early adverse event detection reduce the risk of product downgrade and ultimately improve quality. Adverse event detection. Prediction trending can be used to identify adverse events in the manufacturing process. OPTIX predictions offer insight during process upsets. Additionally, mills can take advantage of the relationships recognized by OPTIX to Operators can use data-driven recommendations to maintain manufacturing conditions or target optimised machine efficiency. WORLD PULP&PAPER 99 Figure 4. (a) Predictive model using poor prediction techniques that oversimplify the process; (b) Predictive model with the same data utilising OPTIX, which is capable of handling paper machine process data.