After the previous texts initially dealt with fundamental questions about Smart Data, Predictive Analytics & Maintenance and the changes brought about by digitization, the last two parts of the blog series deal with concrete application examples.
Predictive maintenance routines can be found not only in industry. Almost everyone has had the annoying experience of standing in front of a defective ATM. Or even worse: the machine retains the card without withdrawing money. So it is not unusual for banks to have to deal with disgruntled customers. All the more reason, of course, for banks to be interested in establishing predictive maintenance and appropriate servicing so that they can be alerted to problems in good time and rectify them. The main problems here too are a parts failure, or that – similar to a printer – the notes get stuck when they are issued and there is a paper jam.
Fixing errors before they occur
In the case of ATMs, new technologies such as the Internet of Things (IoT) allow for more precise monitoring of processes: For example, sensors collect data on all the processes involved in paying out banknotes. This includes data on each successfully completed process, as well as data read by sensors on each individual banknote paid out by the machine, such as: the gaps between the individual banknotes in the bundle, the thickness of the banknotes, or the time interval between the collection and merging of individual banknotes from different stacks. In addition, all maintenance data from previous maintenance is also collected and evaluated. This includes error messages and information on repairs carried out.
Together, all these data allow two models: one relates to the entire withdrawal process, the other to the individual banknotes that are withdrawn. Here too, the aim is to predict possible errors in the system before they occur and minimize customer satisfaction. In this way, a warning system can be established: Based on previous data, the ATM recognizes that a transaction error has occurred, possibly because of paper jams, not all notes can be paid out. The ATM can then cancel the transaction itself and give the customer a warning. Theoretically, it would even be possible for a mechanic on duty to receive a warning directly on his smartphone and carry out a repair.
Predictive maintenance is the best way to quickly and easily take advantage of IoT. This includes, among other things, possible machine failure: Predicting a prolonged machine failure is in the interest of every company, as it can cost tens of thousands of euros per minute. The same applies – and here especially – to intelligent power grids or intelligent transport networks, i.e. sensitive systems. The main goal of many companies is therefore to almost completely exclude a failure.
One example is FANUC, a Japanese manufacturer of industrial automation equipment. For years, the company had problems helping customers optimize their equipment in the various production facilities. Working with Cisco and Rockwell Automation, FANUC then developed a solution that would reduce machine downtime to zero. They do this by analysing data provided by the company’s manufacturing robots: customers share this data and FANUC collects, stores and analyses it in the cloud to identify and resolve potential problems before they can have a negative impact on production.
General Motors, for example, has connected about a quarter of its 30,000 manufacturing robots to this IoT solution, significantly reducing downtime over the last two years. By knowing when a part is likely to fail, the company can initiate timely maintenance and order parts, but does not have to keep them in stock.Predictive maintenance for renewable energies
Predictive maintenance for renewable energies
In recent years, environmental awareness and awareness of renewable energies have grown. As a result, wind turbines are now one of the most important energy sources in the context of the energy turnaround. One of the main components of wind turbines is a generator rotor, which is equipped with a variety of sensors to monitor the condition of the turbines and their status. This data also allows a prediction model to be calculated to determine critical key performance indicators (KPIs). These include the time until a part fails. In one project, for example, data from three different wind turbines at three very different locations were compared with each other – for one year, sensor data was provided every ten seconds on factors such as temperature, speed, turbine output or generator winding. This was used to develop a prediction model, for example, for the life span of generators and temperature sensors. By predicting a possible failure, the mechanics can concentrate on sensitive or suspicious parts during maintenance. The data also allow a better understanding of the causes of part failure.
In all the examples mentioned, one thing is fundamentally true: There are many different solutions for predictive analytics and maintenance based on data and special algorithms. These can be very specifically tailored to individual industry sectors. But what they all have in common is that they collect and evaluate data and thus allow a deeper insight into processes and environmental factors influencing them. This is not magic, but simply statistical models that allow companies to make processes more effective and thus save a lot of money. Even if investments in new technologies are necessary in the beginning.