Presentation on September 28-29, 2017, at the 5th Annual BayesiaLab Conference:
BBN modeling – machine learning derivatives in FMCG, hot rolling mills, polymer extrusion, paper and corrugation industry
The derivatives of the process industries are on multiple domains of thermodynamics, vibration, flow rates, pump dynamics and thermometry in the realm of mechanical engineering; the elements of harmonics profiling, the instrumentation networks, the PLC algorithms, the power profile and finally the quantum of electrical noise in the systems that intervene with the fundamental design of the process and thereby attribute to the structuring of the causal links and the influence nodes in the parametric determinants and finally the key elements in the process engineering models that drive productivity, yield and quality derivatives in the processing.
An integration of multiple domains in the BBN modeling have helped create insights into the process through the invoking of the influence clusters and generating trouble-shooting modes in the process with high degree of real time fidelity of data and resolution of interpretations leading to decision matrices with a phenomenal accuracy rate of 77-80% accuracy bandwidth.
The validation of the BBN derivatives on data drawn in from the case studies has revealed clusters of parametric influence through effective variable nodes. These are revealing to the interested domain experts since in several cases, the dominant domain understanding have been challenged by the results of the BBN models.
The underlying learning from the BBN modeling on a selection of process industries is essentially that of unraveling causal links with a high degree of accuracy; no necessarily in synchronization with the respective industry conventions and therefore has tremendous commercial value in creating detailed process automation logic.