Machine learning is at the heart of many innovative artificial intelligence technologies. Programs developed with ML can predict equipment breakdowns, anticipate customer behavior, and make logical and analytical decisions almost like humans.
We will tell you how companies use machine learning and show examples of using machine learning in real cases.
Machine learning in industry: managing production, minimizing downtime and accidents
Minimizing production downtime. Downtime due to breakdowns, malfunctions or lack of raw materials can cost a plant millions of dollars. Machine learning helps prevent them. To do this, data is collected from the sensors on the equipment, and then they look at what indicators failures occur. In the future, using this information, you can predict when and why a simple one will happen, how to avoid it.
For example, it may turn out that the temperature always rises in the workshop before the equipment breaks down. Then, when the temperature rises, the system will notify the engineers, and they will prevent the problem in time.
To avoid downtime in mining operations, oil and gas equipment manufacturer GE Oil & Gas uses the Industrial Internet of Things and machine learning. The company’s platform collects data on the state of oil production, and then schedules diagnostic checks and helps identify problems before they happen. The same platform helped Kuwait Oil Company increase gas production by 2-5%, and Malaysian oil company Petronas cut maintenance costs by 10%.
Creation of a production management system. With the help of sensors and machine learning, you can not only perform narrow tasks, for example, prevent breakdowns, but also manage the entire production:
- reduce the percentage of defective parts: analyze why a marriage occurs and how to avoid it;
- optimize individual steps so that they take less time;
- use fewer materials for production, which means lower costs;
- monitor the condition of the equipment, record its efficiency and workload;
- automate individual stages of production.
Microcontroller manufacturer Simatic uses an IoT and machine learning platform. It helps to collect and analyze information from sensors on equipment in real time. This helped to automate the production of thousands of types of products by 75%, to increase the volume of production by 9 times with the same areas and personnel, and to reduce rejects by almost 100%.
Identification of security threats. Machine learning helps to make production safer by identifying minor changes in equipment operation and alerting of a possible disaster in time.
For example, the energy company Shell uses machine learning, neural networks and IoT to automatically detect and alert security threats to employees. This is how they manage to react to the problem even before the catastrophe occurs. By the way, Shell also uses machine learning to optimize production and mining.
Exploration of new deposits. One of the main problems in the oil, gas and mining industry is the difficulty in discovering new deposits.
Machine learning helps speed up this process. Based on data from past deposits, artificial intelligence builds models that predict with high accuracy where to look for new deposits of gas or ore.
Gazprom has a Digital Core project . It is a digital laboratory that analyzes reservoir samples using machine learning technologies. The algorithms simulate the conditions where the sample is taken and help create a digital twin of the field. With its help, mineral reserves are estimated and an individual approach to development is selected. This makes it possible to increase the extraction of minerals from a particular deposit by 1.5-2 times, as well as to look for new ones.
Machine learning allows you to create complex visual reservoir models. A source
Machine Learning in Finance: Assessing Risk and Fighting Fraud
Creditworthiness assessment. Usually in banks, the client’s creditworthiness is assessed by managers. Employees spend a lot of time on assessments and often make mistakes – they reject loans to those who could pay them, and give them to those who are insolvent.
The algorithm can be taught to assess the creditworthiness of bank customers. To do this, information about previously issued loans is loaded into it: whether they were paid or not, whether there were delays or early repayment. All this helps the bank to automate the issuance of loans.
For example, Sberbank has created a ” Credit Factory ” – a system that allows you to make decisions about a client’s creditworthiness in a few minutes. In 2020, the bank launched such a factory for legal entities – it helps make decisions on loans for businesses in 7 minutes. Now 98% of loans to individuals and 20% of loans to small and medium-sized businesses are issued automatically, which saves billions of dollars.
Anti-fraud. Banks and their clients regularly lose money due to fraudulent transactions. Machine learning helps to recognize such operations – special algorithms learn to detect signs of fraudulent operations and block them in time.
Many banks have examples of machine learning for fraud prevention. For example, Sberbank uses AI to block suspicious transactions, and recently caught a fraudster with its help. Overseas Danske Bank has reduced the percentage of false allegations of fraud by 60%.
Examples of machine learning in medicine: diagnostics and robotic operations
Improving customer service. The faster the registration process at the clinic goes through, the less queues, the more convenient it is for doctors to work and the more loyal the patients.
.As soon as the patient approaches the counter, the administrator sees the required card on the computer and issues a referral to the required office. This has helped to avoid rush hour queues, simplify the work of administrators and serve more patients.
Diagnosis of diseases. If you load the examination and diagnostic data into the program, it can be taught to make diagnoses in much the same way as doctors do.
For example, Corti’s artificial intelligence listens for ambulance calls and recognizes cardiac arrest based on caller responses, voice and breathing. In one experiment, the program recognized 93.1% of cardiac arrest, while people usually recognize 72.9%. In addition, Corti is faster – making a diagnosis in 48 seconds versus 79 for human dispatchers.
Now the system is being implemented in several European cities – it will work in the rescue service together with dispatchers. The video shows the AI talking to the person who called the ambulance.
Automatic robotic operations. Machine learning is helping to teach medical robots to operate on patients on their own, taking into account many factors.
At the University of California, the robot was “shown” 78 films about operations to teach him how to apply sutures. Thanks to this training, the robot was able to sew up fake wounds, however, with an accuracy of about 85% – this is not enough for real work. Perhaps, in the future, such robots can be used to automate some operations. The video shows the process of training the robot.
Applying Machine Learning to Retail and Marketing: Predicting Buyer Actions and Controlling Product Stocks
Predicting shopper actions, personalized offers and advertising. A trained algorithm can predict customer behavior:
- determine who will make a purchase in the near future;
- understand who prefers which products in order to recommend them;
- offer personalized discounts to drive purchases.
For example, the Rive Gauche cosmetics chain uses machine learning to send personalized offers to customers. The program determines which of the buyers can make a purchase in the next two weeks, what products are best offered to them and what to give out a discount on. The buyers with whom the system worked, the average check was 42% higher, and repeat purchases were 47%.
Uralsib Bank also uses machine learning for personalized offers. For example, the system finds clients for whom the interest on a credit card is not important, but the credit limit and the length of the grace period are important. The bank offers them credit cards with an increased limit. Personalization has already helped boost sales of some products by 25%.
Demand forecasting and procurement automation. Machine learning helps you analyze customer activity and inventory to understand what, when, and how much to buy.
British supermarket chain Morrisons uses machine learning to predict what products will buy and when. The system takes into account many factors such as holidays and weather. As a result, the network was able to reduce supply gaps by 30%.
For more examples of machine learning in retail, see How Machine Learning Boosts Sales .
Machine Learning in Logistics: Conserving Resources and Preventing Supply Disruptions
Save fuel and improve vehicle productivity. Fuel is one of the main cost items in logistics. Machine learning can help you cut costs by optimizing routes or figuring out how to reduce the number of cars while maintaining productivity.
Caterpillar’s Marine Division has implemented machine learning to conserve resources. The company installed sensors on the equipment of the ships and found out that more generators at lower power work more efficiently than the maximum use of several generators. This solution saved over $ 650,000 in a year.
Prevention of supply disruptions. Delaying even one vehicle leads to disruption throughout the supply chain: downtime, loss of money and customer dissatisfaction. Machine learning helps to avoid this: predicts risks, helps to prevent them in time and adjust delivery times, taking into account all factors.
DHL uses artificial intelligence on its Supply Watch. It monitors various risks: weather conditions, environmental factors, traffic congestion and even crime rates, in order to inform customers in advance about possible delays in deliveries.
Prospects for the application of machine learning
In 2020, 34% of companies in Europe, the United States and China are using artificial intelligence and machine learning. According to experts, the machine learning market will grow by 42% by 2024 .
According to a survey by Algorithmia , in 2020, companies are most likely to use machine learning to cut costs, better understand customer behavior, and improve customer service.