Читать книгу Profit Maximization Techniques for Operating Chemical Plants - Sandip K. Lahiri - Страница 46

One of the major profit suckers in chemical industries is a sudden trip of critical single line equipment. Once a plant trips, millions of dollars get lost in terms of less or no production and more time is required to bring back the plant to on‐spec production after a disturbance. Besides this, a lot of money is lost in terms of flaring, venting, or draining of costly chemical gas or liquids (Wang, 1999).

Big data analytics can be used to develop fault diagnosis software to anticipate the failure of critical equipment at a very early stage and thus give sufficient time to plant engineers to take preventive or corrective actions. Such fault diagnosis systems analyze historical data to generate insights that cannot be observed using conventional techniques. By implementing an intelligent analytics‐based fault diagnosis system, companies can determine the circumstances that tend to cause a machine to break. Then a real‐time automated system can be developed to monitor all relevant parameters and give early fault signals, so engineers can intervene before breakage happens, or be ready to replace a component when it does, and thus minimize downtime. Companies who has implemented such systems typically reduce machine downtime by 30 to 50% and increases machine life by 20 to 40% (Wang, 1999).

Chemical companies are already starting to see substantial gains in this area. One major polymer producer consistently ran into problems with extruders at its largest plant. When one of the shafts of the extruder broke, the plant had to stop production for 3 days while a replacement was installed; these shafts are expensive, besides the cost impact of the production loss. Engineers had done a detailed study to determine the possible root causes of failure; alternative materials in the shaft were also tried out, as well as different process conditions, but none of them solved the problem.

A principal component‐based fault diagnosis approach changed all this. It combined a detailed analysis of data from hundreds of sensors with the plant engineers' expert domain knowledge, and reexamined the process variables and other data sources; it then developed a real‐time‐based algorithm to predict when a failure was imminent. The problem occurred with only one of the polymer grades, and not with all batches, suggesting the key lay in specific process conditions in the equipment. The team developed a model based on “a hybrid principal component analysis and artificial neural network” algorithm that took into account the specific parameter settings in production, such as extremes of temperature and temperature progression, together with information on the polymer product type and composition.

A real‐time visual platform was developed, which flagged an early warning to engineers when the plant conditions approached a state that could ultimately lead to shaft failure.

When it flags that a failure is imminent, the plant operators undertake a 15‐minute cleaning of specific parts of the machinery to prevent the failure from occurring. The improvements to performance that resulted from using the advanced‐analytics approach have been substantial. Instead of a 3‐day production loss plus a costly extruder shaft replacement, the company was now dealing with just a 15‐minute production interruption, and the approach has cut production losses by 60% and maintenance costs by 85%.