This article focusses on time-variant manufacturing processes, and how analytics can help boost their productivity. Many production lines relying on such processes are involuntarily operated using a suboptimal combination of input control settings most of the time. This weakens overall returns at facilities such as steel mills, mines, chemical and other manufacturing plants.
Heavy manufacturing largely relies on two types of processes: structured, deterministic processes and less structured processes.
In structured, deterministic processes well-established thermodynamic and kinetic engineering principles precisely describe plant operations. The deterministic nature of these type of processes enables manufacturers to accurately set input controls – like flow and oxygenation rates, heating intensity and engine torques – to maximize an asset’s productivity.
Less structured processes tend to constantly deviate from their theoretically optimal state due to small, real-time variations in process flows and external conditions. For example, most chemical processes fall into this category. They are often affected by external factors like weather and the exact chemical consistency of the inputs being fed into the process. In steel making, the complex interplay of many individual components in the production line has a direct influence on the quality of the output. The compounding and time varying influence of these factors makes it difficult to accurately steer the results of these processes based on engineering principles alone.
Addressing issues with less structured processes has proven challenging. Attempts have been made through implementation of advanced-process-control (APC) systems. While being very powerful for well-described (or “deterministic”) problems, the results have proven less satisfactory in the context of less structured processes. The main reason for this is that APC systems tend to optimize only one subcomponent of the process. This is less effective than performing an integrated optimization of all factors in the end-to-end process.
Artificial intelligence and machine learning provide new levels of insight
The emergence of IoT technology – which includes inexpensive sensors and reliable data transmission combined with affordable storage and powerful and easy-to-use analytical platforms – has made it possible to provide new levels of insight through deployment of artificial intelligence and machine learning approaches.
In a first step, a digital version of the industrial process is built using machine learning techniques. This digital copy – sometimes called a digital twin – captures the relationship between sensor readings (e.g., temperature, pressure), input control setting (e.g., valve opening rates, heating intensity) and asset productivity. It explains the complex and time-different variability of the industrial process beyond what is captured by theoretical engineering formulas. The reason for this is that is based on the exact features the real-life process rather than a theoretical approximation.
In a second step, optimization algorithms are deployed to fine-tune all input control settings simultaneously. The digital twin approach allows this to be done continuously and in real-time. It thus ensures that a production line’s input controls are always set to operate at peak efficiency.
Substantial improvements by digital twin approach
The results of this approach have proven dramatic. Substantial improvements in overall productivity have been achieved. One chemical manufacturer has managed to increase the throughput of one of their major production lines by nearly 20%, while simultaneously increasing the proportion of useable end-product by 6%. In another example, a gold miner now manages to extract 4% more gold from the same ore in one of its chemical extraction processes. Early adopters in the metal production industry have similarly seen major benefits, with one steel producer seeing an increase of 14% in first-time-right production. Looking beyond manufacturing, internet giant Google has achieved reductions of up to 40% by deploying this type of machine learning approach to reduce the energy used for cooling its data centers.
By increasing process efficiency, the analytical digital twin approach can greatly improve performance while reducing costs. Strategy consultant McKinsey & Company estimates that the value can be worth tens or even hundreds of millions of Euros when manufacturers systematically apply these techniques to all individual production assets within their organization.
Discover more insights on SAS Industrial IoT Information Hub