Digital enterprises are different, but they all share one important feature: they streamline their daily processes and make operational decisions based solely on data—and they use the latest version of data. Intuition or preliminary plans play the second fiddle.
In simple situations, information can be analyzed mentally or in Excel spreadsheets. However, as time goes by, the systems become more complex, data volumes burst, and more process parameters should be considered. Digital companies use trained mathematical models to analyze data.
With these models, Softline Digital customers reduce costs, perform quality assurance, eliminate unscheduled downtime, and solve many other problems. Here are a few examples from our practice in the oil and gas and nuclear industries, as well as in government agencies.
1. Express control of gasoline parameters at a refinery. Virtual sensor.
Product (e.g., gasoline) quality control is traditionally performed at oil refineries by standard laboratory methods for physical and chemical properties measurement regulated by the relevant documents (state standards (GOST), technical specifications (TU), enterprise standards (STP), contract conditions). This kind of analysis brings a result only after 1 to 3 hours. This means that in case of non-conformance, the refinery will deliver low-grade products for another 1 to 3 hours.
With the automatic online analyzers that are part of our solution, oil refineries can continuously control gasoline quality parameters (as well as parameters of other petroleum products) in real-time mode based on infrared spectroscopy data (octane number, saturated vapor pressure, friction composition). The developed mathematical model describes the dependence of gasoline characteristics on the quality of feedstock and specified equipment parameters. If the quality of the gasoline differs from the expected value, the technological process has to be adjusted. Most importantly, all problems are reported without delay.
2. Reliability assessment of oil well pumping equipment
One of our customers, a Russian oil company, needed to predict failures of electric drive centrifugal pump and other equipment for oil extraction from wells. The solution had to take into account different operation modes and geological conditions, and help plan preventive maintenance.
The customer provided extensive historical data about the failures of various units, which made it possible to build an operational model of pumping equipment in oil wells. The model can determine the optimal equipment configuration (combination of specific unit and device models) for any specific oil well and thus maximize the mean time between failures.
The solution reduces operating and capital costs by minimizing the number of incidents that require on-site visits of repair teams, which turn out to be very expensive due because oil wells are often remote and difficult to reach.
3. Forecasting electricity consumption volumes and tariffs
Large enterprises consume a lot of electricity. They must forecast consumption and meet this forecast. Compliance with the plan allows to choose optimal electricity tariffs, and non-compliance causes excessive costs.
The customer has set Softline Digital the task of optimizing electricity costs through accurate hourly consumption planning with an error margin of no more than 1.5%. To solve the problem, we preprocessed the data, seeking, excluding, and smoothing anomalies and gaps. Then we built predictive models for time series and dashboard data, taking into account various factors, such as weather, calendars (production, workloads, maintenance works), and macroeconomic indicators.
4. System for technical diagnostics of equipment at nuclear power plants
Although the nuclear power industry has effective means of preventing emergencies and monitoring equipment health, minor equipment failures are not impossible. Our customer, a nuclear power plant, accumulated an exhaustive database of such failures and the results of their analysis. The customer representatives asked Softline Digital to create an expert system that would enable quick monitoring to identify and classify failures based on their characteristics.
Having created the model, we trained it on the customer's historical failure data. The model became the core of a monitoring and alert system that tracks abnormal situations and helps personnel classify them based on a statistical analysis of the failure archive database.
5. Q&A system
According to the survey conducted by the TAdviser analytical agency together with Naumen in Q1 2021, about half of Russian businesses (large industrial enterprises and upper-middle businesses) faced problems in 2020 when employees were searching for the necessary information to solve business tasks. Sometimes you need to get the information quickly, but cutting through the clutter of different documents and messages and combining the discovered data requires a lot of time.
To solve the problem, Softline Digital specialists have implemented a prototype of a smart search system. Any employee can ask it a question, and the system finds the most relevant documents and gives a quote as an answer.
6. Rate setting for the stationery market
Softline Digital team has created a deep analytics tool, Digital Auditor, for the Accounts Chamber of the Russian Federation. It allows the institution to monitor purchases, compare prices, and perform visual analysis via dashboards.
Before the project, there were no price thresholds for many goods, works, and services purchased by state organizations, as well as guidelines for calculating the allowable prices and detecting the excesses. As a result, there were many cases when at the stage of budget review, the planned budgets were found to be overvalued.
In the "Digital Auditor" system, each product is described by a probability distribution of price. It assesses the distribution and derives its parameters (type, expected value, variance) to determine the recommended (fair) price and the deviation between the declared and the recommended one.
After calculating the fair price, the model calculates the difference between the fair price and the price for which the goods have been purchased, and assesses how significant the difference is. Softline Digital also developed an index to rank organizations on the level of deviation from the fair price, taking into account the purchase volume.
7. Tender procurement monitoring
This is an internal project of Softline. The digital tender site had limited search capabilities, and the evaluation of the tender participation feasibility was too long and difficult. The solution was a product for automated tender search and evaluation.
To train the model, we took all tenders from one public procurement site over a certain period and divided them into "good" (experts from the tender department manually classified them as good, started working on them and added them to the CRM) and "bad" (those that were not added into CRM). The trained model was able to distinguish "good" tenders from "bad" tenders with 98.5% accuracy!
In other words, the work that used to be done manually is now done by the model automatically. As a result, search costs have decreased and the speed of finding potentially interesting tenders has grown. The tender department staff has now switched from routine tender monitoring to more intellectual work and tasks that require expertise.
Instead of a conclusion
Data analytics from Softline Digital excel at creating solutions related to predictive analytics, machine learning, and big data. The solutions that we've told about are just a few applications of data analysis. We invite you to discuss how analytics can solve your business challenges on a brand new level