An Italian company is working in oil refining, providing process design studies and technical assistance aimed at maximizing operation of process units, considering both energy performance and production yields. With experience in continuous chemical processes, typical of oil refining, petrochemical or chemical industry, ti disposes of good chemical process knowledge and comprehension capability.
Modern chemical processes include sensors constantly producing a considerable amount of data. In addition to data automatically captured by Distributed Control Systems, datasets representing process operation can include laboratory tests for quality control or other characteristic process information.
Data monitored by plant operators are typically automatically stored in historical archives. Sometimes data are used for production monitoring but, in general, they are not much used to improve plant operating performance. The main obstacles to the systematic application of data for the improvement of plant operations are the following:
• Identification of meaningful variables: monitoring a chemical process requires installing a large number of sensors to detect the status of the plant at key process nodes. It is not easy to identify the subset of variables to consider to solve a problem.
• Variable correlation: in case of a process plant, values of variables are highly correlated but it's difficult to exploit this feature systematically and organically to define status of the plant and identify any anomalies.
• Data organisation: data representing the state of a process are sometimes heterogeneous, sampled at different frequencies and produced by diverse sources (e.g. laboratory analytical data, process variable values, feedstock quality). Effective use of data requires pre-processing to generate congruent and representative data sets.
• Data reliability: measured values are not always correct and it is not easy to identify reliable values.
• Variable multiplicity: state of a plant is represented by many variables that the observer should visualise in a holistic and aggregated way. Typical method of analysing one variable at a time or the correlations between two variables limits the understanding the process status.
Proposal aims to investigate application of chemometric methods (i.e. multivariate analysis, experimental design, data models) to study data from chemical processes to extract helpful info for improving plant performance. Possible fields are:
• Process monitoring support: providing operators with aggregated views of variables helpful in predicting the evolution of plant towards sub-optimal operating states.
• Prevention of undesirable conditions: study of critical situations and understanding causes that generated them to prevent them in future.
• Identification of malfunctioning sensors: multivariate analysis for identification of uncorrelated outliers.
• Soft sensors: use process data to develop inferential sensors predicting an unknown variable as a function of other measured variables.
To achieve desired results, it is essential to place statistical analysis tools offered by chemometrics in hands of technicians with in-depth process knowledge who can correctly interpret results and convert them into information or applications useful for process optimisation. The goal is to apply these technologies to solve real industrial problems. The result is developing practices to provide specialised data-driven chemical process troubleshooting consulting.
The consortium consists of an Italian SME, a chemical industry as industrial tester and a Spanish university as scientific partner. Looking for industrial partners to apply these methodologies to solve specific operational problems, providing plant data and working to improve performance or prevent undesirable conditions.
Call: HEUROPE -Sustainable, secure and competitive energy supply -deadline: 27 Oct 2022
EOI deadline: 30 September 2022