Data in Industry: How to Collect and Capitalize on It

It's nice to read the news: researches show significant industry successes in digital transformation* (KMDA, 2020), and companies like SIBUR show cases of quick and noticeable business impact from data analytics in production. However, working on projects in different sectors, we can see that industrial companies are not yet in general taking advantage of all the business development opportunities offered by data.

Competitive advantages and benefits seem to be obvious: predictive unscheduled equipment shutdowns, cost reduction through process optimization, and higher quality and quantity of products. So what's stopping industrial enterprises?

The fact is that it is more difficult for them to get the data collected and analyzed than for other companies.

1. We need to digitize analog data from equipment and machines.
2. These enterprises have a huge amount of information - petabytes of data per year.
3. There is a shortage of people with relevant experience: so far there have been few such tasks, and thus, the industrial sectors - oil and gas, metallurgy, energy, machine building - are at the outset of their way.

However, it is noticeable everyone has already realized that the use of data in operation is inevitable, and it gives a competitive advantage.

New people and businesses - even those with less industrial experience but with data know how - are starting to crowd out the incumbents. It seems that all companies willing to stay in the market will have to collect and analyze data in the near future.

What data and how to collect in the industry?

It is very expensive to equip everything with sensors. There are many links between the equipment and the digital monitoring system. It takes an entire infrastructure to gather information:

• analog sensor,
• logic software controller, oscilloscope or other device to digitize information,
• transmitter that sends data to the repository,
• data repository.

Moreover, the more links in a chain, the more potential problems in data analysis:

• measurement inaccuracies,
• process limitations,
• fluctuations through external factors,
• digitization errors,
• electronics that occasionally malfunction.

Companies sometimes outfit equipment with million-dollar sensors and don't know then what to do with the data, or the optimization they do with them doesn't pay off.

What to collect? There is no universal answer to this question. Everything depends on the equipment, on its operation mode, on the specifics of production processes, on developed and proven models, and in each case, it is a matter of design. At the design stage, the chances of making mistakes are much higher if you are inexperienced. Where to place the sensors, at what frequency to measure, how to interpret the data - all this gets right only by experience.

The business analyst decides what kind of data and from what equipment is needed to be analyzed.

The main sources of data are sensors, video cameras, manual input systems, and incident logs maintained in production. The business analyst decides which data from these sources is worth collecting and analyzing. This is an expert who understands how the production process and equipment is arranged, which parts it includes, and how they are related to each other in terms of logic, mechanics, thermodynamics, etc. The expert indicates which of the measured parameters most clearly reflect the equipment operation and the processes in place.

The complexity of data collection usually depends on the production conditions - high temperatures, dust, etc. - and on how the equipment is fitted with the required sensors. In most cases, retrofitting is associated with interfering with the device design, which means a loss of warranty, risk to the function and safety of the expensive equipment. There is a non-invasive retrofitting, however, the accuracy becomes here an issue, cause there are units from which the data simply cannot be extracted this way: for example, from the internal insulation in electric motors.

The collected data needs to be cleaned, i.e. the unnecessary information has to be deleted. The data is often cleaned by simply cutting off threshold values. However, Ctrl2GO Solutions takes a different approach: we believe in eliminating possible errors through smart algorithms tracking which sensor may have failed or is showing incorrect values.

What to do with the collected data

It is possible to collect a huge amount of data on the equipment operation by recording certain parameters within a production cycle. However, all this information without analytics will be almost useless: it will not provide any discoveries by itself.

Data scientist is the king here. Depending on the task and the amount of data, he determines which processing algorithms can be applied and in what sequence, and what results they can lead to. Usually, several hypotheses are developed and then tested. The proved hypothesis usually becomes a solution to the problem - an analytical model that is put into industrial use.

A data analyst is someone who derives from collected data a practical value for production

If a company has several production facilities with the same equipment, it is better to collect data on all units. A machine or pump may have several dozen of the most likely failures, but it is unlikely that they will all happen to one particular piece of equipment. Therefore, the more data the model receives for training, the more accurate it will be.

The created model analyzes data on equipment operation, information on ongoing production processes and the influence of external factors and identifies anomalies. Thus, the cause of a failure can be found that has already occurred - for example, that's how we at Ctrl2GO Solutions used historical data to find a fault in the turbine generator of a heat power plant, which caused an accident only a year and a half later.

Once trained on historical data, the ML model can not only monitor the equipment condition in real time, but also perform predictive analytics, it means it can predict in advance the time and probability of failure on a component or unit. It is also possible to calculate a "health index" of the equipment, which allows to move from scheduled repairs to condition-based repairs.

The data path in industrial production: from sensors on equipment to finished reports on operators' displays.

Why your business needs data

The right approach to using plant and production data can help industries achieve real business results - both operational and strategic ones.

Optimizing processes. Real-time data use allows to make necessary adjustments to production processes - for example, to adjust the operating modes of a flotation plant at a ore refinery, which will reduce reagent consumption and improve recovery.

Higher performance. The data on raw material quality helps a metallurgical enterprise to ensure the correct ratio of concentrates during charge production and thus to increase the performance of furnaces by 2-5% and to improve the quality of the output product by 1-6%.

Reduced machinery and equipment downtime. Telemetry data from diesel locomotives will help to reduce downtime for service by 12%, rolling stock diagnostic time by 4 times and maintenance and repair costs.

As other industries show, data analytics can improve management efficiency and identify business bottlenecks. However, there are strategic benefits that are specific to industrial plants - for example, extending the life of existing equipment while keeping it running at peak performance and reducing environmental risks.