Let's start with a question: is the term Industry 4.0 a blessing or a curse on the road to networked production?
Papenfuss: With Industry 4.0, we and companies are seeing a huge change coming our way. The term encompasses a wide variety of trends. While digitalization used to be a marginal topic for many companies that only a few plant managers and engineers were concerned with, it has now reached the very top of management for the first time: Whether you're talking about Industry 4.0, digital transformation or the Internet of Things, in the last two to three years the development has been driven by top management. This has put more pressure on companies to really get to grips with the topic of Industry 4.0. New interdisciplinary teams are developing and budgets are increasing. We at OSIsoft have been dealing with data and process analysis for 38 years, especially in the process industry, but now more and more companies are getting involved in the topic and bringing their perspective to the table. This sometimes causes confusion for users.

How are you and your customers reacting to this?
Papenfuss: Suddenly, traditional IT companies such as IBM, SAP and Microsoft are getting involved in this segment and investing heavily. However, they often still lack the engineering knowledge, i.e. the background knowledge of the respective industry. The confusion among customers is simply higher and therefore the need for explanation increases. On the other hand, there are the positive aspects that this development brings with it: the greater importance in companies and the support from management also increases the budget available. Another advantage for us is the increasing competitive pressure, which also increases our innovative strength. It's never healthy to be the only one on the market.

Was OSIsoft ever completely alone on the market?
Papenfuss: We have always had many unique selling points and are simply regarded as the benchmark in some sectors. But there have always been competitors for special niche applications in a wide variety of sub-sectors.

Which were they?
Papenfuss: For example, providers in the metals and mining sector who offer special solutions, including video and audio data. These are solutions that we do not offer and that cannot be compared with our PI system. Just like a car and a motorcycle cannot be compared: You can use both to get to your destination, but in different ways.

Do you have customers who have never heard of digital transformation?
Papenfuss: Surprisingly, yes! There are still many companies where work is still done manually or where there is still a lot of paper in production.

How do you explain to them that they need to go digital?
Papenfuss: There are various starting points. Of course, a lot is happening here as a result of the Industry 4.0 trend: management is dealing with it and this is shaking up the employees. Another approach is the typical sales approach: we show where there is potential for savings in the company. Thanks to our many years of experience, we have many examples that we can use to show potential customers how much money they can save through which optimizations.

In which industries are you active?
Papenfuss: We are mainly active in the process industry. We started out in the oil and gas industry. It quickly became clear that the solution could also be used for other companies. Although oil and gas is still one of our strongest segments, we are also active in the energy, chemical, pharmaceutical, paper and food industries. We are not yet widespread in discrete manufacturing for two reasons: Firstly, the discrete industry is not yet as hungry for data as the process industry, and secondly, we have lacked the focus and also the capacity to date. In the last six years, our workforce has grown from just over 700 to around 1,400 employees.

How and where do you recruit your specialists?
Papenfuss: We are well known in our field. Many engineers have been familiar with our PI system for many years and have already worked with it. That is our advantage: when these people switch to us, they already know what possibilities the software offers. We also recruit specialists directly from university at a very early stage: graduates usually start in support and can work their way up into a wide variety of functions within the company. And then there are career changers like me. However, every employee is a specialist in their field.

The PI system is complex and certainly requires a lot of specialist knowledge...
Papenfuss: Yes, above all engineering knowledge is required here. We monitor the entire process chain. The required sensor data is either generated by an IoT platform or provided by a control or SCADA system. Our PI system manages the entire process data: from data acquisition to archiving, real-time analysis and forwarding the data to the application or specialist who needs it. This requires a great deal of specialist and process knowledge: from the source system to the general nature of the sensor data, which is fundamentally different from relational or transactional data in IT.

What applications is the PI system used for?
Papenfuss: Our customers use the PI system for energy management as well as to improve product quality or to ensure regulatory compliance, for example in the pharmaceutical industry, which requires seamless recording of the production process. The data that we collect in a plant is analyzed in real time and correlations are determined with the help of algorithms. In this way, maintenance intervals can be predicted. However, predictive maintenance is only one possible application.

But there are hundreds, if not millions of sensors in a system. How do you select the right data here?
Papenfuss: Of course we are talking about large amounts of data here. Our customers, for example a chemical company, have well over 300, up to several million data points in our system, i.e. different measuring points. Power plants have up to 10,000 data streams to monitor a large turbine. A wind turbine has between 300 and 400 data points, which are read out every second or millisecond. The only question is which data is important. In general, however, all data is stored because you never know what you might need it for later.

How large is the amount of data then?
Papenfuss: The PI system is not a traditional database, but rather software that manages data. The data volumes are therefore relatively small, we are talking about giga and terabytes here.

Back to the wind turbine with 400 data points: How big is the file if the turbine runs for a day?
Papenfuss: We are talking about the upper kilobyte range up to a single-digit megabyte range (depending on the data frequency of the source signals). A sensor data tag is made up of an ID, a value, a time stamp and special quality features. If a value does not change over a certain period of time, only the start and end points are saved. You can also set that only larger or relevant changes are saved. For example, sometimes only one or two values are saved per hour.

Suppose a production company with several machine tools approaches you and wants to optimize its plant. How do you proceed?
Papenfuss: I'll start by saying that we don't optimize systems. Our software makes it possible to process sensor data in real time and make it available to the operator for a decision. It's similar to the speedometer in a car: it shows me in real time whether I'm going too fast. If I am, I initiate countermeasures - I brake. It's similar in a power plant: the operators have to keep an eye on the condition of the turbine. If the pressure or temperature is too high, the data is forwarded and countermeasures are taken.

However, if the customer wants to optimize a process or introduce predictive maintenance, we can support them. To do this, we first need to know how we can acquire the data. That's not so easy, because there are a wide variety of communication standards in the industrial sector. In addition, although many manufacturers support a standard, they adapt it to their application. This means that although the systems officially speak one language, they speak different dialects. As with us: we speak German, but the North Germans don't understand the Bavarians. So first we have to understand which interfaces are available. To do this, we draw on our library of over 400 standard interfaces and then install software that picks up the data and transfers it to our system.

So you use the existing sensors in a system.
Papenfuss: Yes, we use the existing sensors and read them out.

Is the data from the existing sensors usually sufficient or are you reaching your limits here?
Papenfuss: In principle, the existing sensors are sufficient, because we can read out everything that has a time stamp and a value. The question is: what does the customer want to do? We certainly have operations that are not automated. In other words, there are not enough sensors available to allow our software to be used effectively. In such cases, we are simply too expensive and the effort is not worth it. The degree of automation is therefore crucial. I don't buy a Ferrari to drive to the shops.

When you have a new customer, do they usually have historical data that you can fall back on?
Papenfuss: In the pharmaceutical industry, yes, because there is a documentation obligation here. In the process industry too, as a rule, but there are many small companies whose data is available in a wide variety of formats. This means that we often have to struggle to get the data out of outdated systems, some of whose manufacturers or developers are no longer available.

How much time do you need to evaluate this historical data?
Papenfuss: It usually takes no more than two weeks for the old data to be transferred. However, there are also exotic systems in which the data has to be loaded manually and this can take time.

Now you have the data, how do you decide which data is relevant?
Papenfuss: In most cases, it is important to include all the data, because we never know what we might need it for. With machine learning and artificial intelligence, we have the opportunity to recognize correlations that we couldn't see before. And, of course, more data helps. We are also now able to use new algorithms that make previously useless sensor data usable. Weather data is a good example: In the renewable energy sector, it is used to make predictions about the productivity of a plant. It was relatively useless 15 to 20 years ago, but now we can use this data to recognize certain correlations. We therefore recommend recording as much data as possible.

As a rule, however, it is still the human being who decides which data is relevant. A machine operator or even an engineer in the process industry knows which data is relevant and which data is missing in order to achieve the desired result. However, certain dependencies can also be identified using data science, which also makes the data relevant.

In production, an experienced machine operator can hear from the noise that something is wrong with the machine. How do you map this know-how in your software?
Papenfuss: This is indeed an area that is currently being commercialized by some companies. We have been working with an Austrian company that solves precisely this problem for several years. One example: In one plant, there were filters that were difficult to access and clogged easily. In the past, a technician had to come and listen to see if a specific noise was occurring. A microphone was therefore installed to record the noise and store it in our system. An algorithm was then developed in collaboration with the technician so that the sound data can now be used to predict when a filter is clogged. By linking these findings with the control system, it is even possible to predict which filter is blocked.

We are currently working with a large number of mechanical engineers who are packaging their knowledge into such algorithms. New sensors, such as microphones or thermal imaging processes, provide the necessary data.

I would like to come back to interoperability. Do you sometimes reach your limits in terms of standard interfaces?
Papenfuss: Of course we always reach our limits, as new standards are constantly coming onto the market. Our strength is that we have a team with many years of experience that only deals with interfaces.

What are the biggest challenges in networked production?
Papenfuss: The central question is how do systems communicate with each other? In industry, there are not just machines and control systems from one company, but a colorful hodgepodge of different manufacturers. They differ in their degree of automation and communication standards. The task now is to bring them together. With our PI system, we build a bridge between the different platforms and thus provide an overall view of the process.

What do you think of communication standards such as OPC UA and TSN, which are intended to ensure interoperability across all systems?
Papenfuss: That's good in principle. OPC UA is seen as a panacea, but it has yet to reach production. In addition, as already mentioned, there are many different "dialects": With every customer, you encounter a different one and have to start from scratch and get in touch with the manufacturer.

What advantages does Industry 4.0 bring?
Papenfuss: Industry 4.0 ensures that the teams are mixed. When I started at OSIsoft six years ago, 85% of the team members were engineers. Today, the proportion of computer scientists has increased to around 30-40%. Engineers are OT experts: they know the control system and the data very well and know what they can do with the data. But until now, the world of what else can be done with the data has remained closed to them. An engineer sees the data primarily in a time curve, while a computer scientist sees the data multidimensionally, i.e. from different angles, due to their experience with business analytics or business intelligence. In IT, there are tools that OT has not yet used. In my opinion, the current situation is very good, because both areas can mutually stimulate each other's work.

How does the link between OT and IT, engineers and IT specialists work?
Papenfuss: It's interesting, because the two areas first have to come together. Engineers sometimes lack the imagination to see what can be done with data. Machine learning, for example, is absolutely amazing: some customers have data that is 20 to 30 years old and suddenly we can recognize correlations and optimize manual processes with the help of machine learning.

What have you personally learned from working with engineers?
Papenfuss: Engineers have an incredible amount of expertise in the field of sensor data and enormous knowledge of the process chain. They know exactly how a system works. I personally have learned a lot about the technical interrelationships.

You cooperate with companies such as Amazon and Microsoft. What exactly does the collaboration look like?
Papenfuss: As a rule, our system is installed close to production. In principle, our partnership with Amazon is again about IT and OT: thanks to the cooperation, we now have the option of sending the data from our PI system to the Amazon or Microsoft Azure cloud. The data is still with us, but the calculations are performed in the cloud as it offers greater computing power. This is very German: the data remains with us, but can be uploaded to the cloud for a short time for analysis purposes.

Is it to do with security or is it a real problem or is it simply fear?
Papenfuss: It is a real security problem, yes. But you have to look at it in a more differentiated way: Our PI system is often used as a critical system, in other words: if it fails, the lights go out. That's why the software has to be as close as possible, i.e. in the network. Of course, you could install the entire system in the cloud, but the question is: is it close enough to real time and what happens if the connection is lost? To put it bluntly: If our software is located in the cloud, the probability of failures due to the infrastructure is many times higher.

One security aspect is also: American or European cloud. The customer decides where the data is located. Many customers do not want their data to end up in America. In the chemical industry in particular, data on substances that could be used for weapons must not be stored outside Europe. This means that certain chemical companies that manufacture in Germany have to take meticulous care to ensure that not a single value ends up in America. But these are legal issues with which we are confronted, but which ultimately do not affect us.

Where do you see the biggest challenges for Germany?
Papenfuss: The will to do something! We are seeing a lot of resistance at the moment, especially from SMEs: "We have always done it this way and we are the market leader, so it will work in the future too" is the motto here. This attitude is certainly also due to the fact that we are doing very well. As a result, there is very little willingness to change anything. And confidence in the new technologies also has to grow first.