A lot has been said about artificial intelligence over the last few years, but do people know how to use it efficiently? The most useful area of artificial intelligence, machine learning, enables us to mine useful data from challenging unstructured data sources typically composed of text, sound and images. Successful data mining produces structured data that can be used, either alone or combined with other indicators, to control a specific application and create reports on it.
Unstructured data – untapped source of data in the world of digitalisation
How many factories or system vendors use artificial intelligence to solve problems in automation or control systems? Generally, the log data and messaging traffic resulting from a process can amount to several gigabytes even during a short period. A manual analysis of masses of data is time consuming and finding the root cause may prove impossible in situations where the problem occurs rarely or occasionally.
With artificial intelligence models, the real-time analysis of event logs and detection of anomalies can be done quickly and easily for specific periods of time. When precise dates and times as well as abnormal log entries are determined automatically, the work of system specialists will also become more efficient. Progress is quick because abnormalities can be detected within several periods of time, such as hours, days, weeks, months and years.
One of the success stories of deep learning in 2017 was when speech recognition achieved human capability. What makes this achievement remarkable, is the fact that the same machine learning model can also be used offline. The model can be integrated into machines, wearable devices and the work environment, allowing people to use their hands to do the actual work. In addition, sound can be used in various ways. One of the simplest applications involves using the frequency of sound to predict the maintenance of machines.
Deep neural networks also allow for automatic screening of image and video material. What makes this technology interesting is the fact that the video material can be processed with high accuracy and in real time. The analysis produces reference points for objects and coordinates detected in images. If needed, the emotional state of people can also be studied.
A cloud connection is not required for the analysis since the neural network is run locally using the principle of edge computing. Privacy protection can be ensured by destroying the recorded image after analysis since there is no need to store it. Typically, only the results of the analysis are sent to the cloud.
Real-time video analysis is an excellent tool for analysing human flows, monitoring work sites, checking the suitability of equipment, optimising the effectiveness of shop window displays, business centres and real properties, as well as for assessing user experience. In the manufacturing industry, it can be used to facilitate the quality control of products. Many processes are still dependent on visual inspection of product quality. With artificial intelligence, the visual quality of products can be controlled through self-directed processes and, if necessary, an entire production line can be stopped to minimise wastage.
Artificial intelligence is not enough in itself
The statement "Artificial intelligence is a good servant but a bad master" is a good analogy of how artificial intelligence should be looked at and, more importantly, implemented. Artificial intelligence as such does not solve anything. Artificial intelligence provides the best benefit when it is integrated into a system that allows continuous training of artificial intelligence as well as visualisation and comprehension of predictions.
Fundamentally, the added value from artificial intelligence is based on its ability to generate accurate predictions from hundreds or even thousands of input variables. For example, a neural network that identifies objects through deep learning, uses pixels as input to produce a prediction of the probability that specific objects will be found in an image.
Predictions and decision trees serve as tools for decision-making. Artificial intelligence learns to predict users’ behavioural patterns based on given sample decisions. When a model generated with artificial intelligence becomes sufficiently accurate, it can also be used to automate real-world operations.
The accuracy required of a model is determined on a case-by-case basis. For example, if the aim is to control a light switch through video data, automation is not a problem since the impact of an incorrect decision is negligible. On the other hand, if an individual malfunction can pose a hazard to humans or cause significant costs, the accuracy of the prediction should be much higher before automation can take place. Business calculations and risk assessments should always be made before trying to increase the level of automation.
Open interfaces and standardisation are an essential part of successful integration. In industry, the importance of a common interface for devices has been understood for decades. OPC UA is one model student worth mentioning. With the successful introduction of artificial intelligence solutions, Wapice’s extensive technology expertise at all levels of the system hierarchy, from embedded systems to cloud solutions, combined with data analysis has proven to be an excellent mix.
Wapice uses functions based on artificial intelligence in connection with sales systems, IoT systems and customised client systems. Artificial intelligence also helps in programming when used in connection with software development systems and testing systems. It is no exaggeration to say that artificial intelligence will be an important element in nearly all programmable systems in the near future.
The original text was published in the Partner Blogs section of the Tekniikka ja Talous magazine on 25 October.