Home » Technologies & Materials » The Road to Digitized PM Production

The Road to Digitized PM Production

Digital technologies and data-based methods will find their way into all industries and ultimately cannot be prevented. The remaining question is how companies want to deal with digital transformation in their business processes and how quickly a recognisable benefit can be derived. Based on a typical PM or advanced ceramic parts production, this essay wants to give impulses, which potentials exist through intelligent data mining at connected powder presses, in order to increase the production performance, to monitor the quality and to reduce manufacturing costs. The focus is on the provision of qualified, refined data for production control, quality assurance and maintenance with the aim of being able to make fact-based decisions quickly and reliably. Full transparency of what is happening on which machine at any time in the production environment thus creates the conditions for comprehensive process control and provides the basis for the digitized Smart Factory of the future. 

1 How to get to a digitalized PM production?

Everyone talks about digitization (the emerging technologies in the Gartner Hype Cycle [1]), but the impression is that many people are consciously or unconsciously talking at cross purposes. Often there is also a lack of basic understanding of the possibilities and limitations of digital data processing. This perception from many conversations with industry partners and customers shows the dilemma that the digital transformation is currently facing in practice. Of course, most of us will agree that the digitization of industrial processes is reasonable, necessary, and beneficial. But when it comes to concrete approaches to digitization in the company’s own manufacturing sector, reasons are immediately put on the table as to why this can hardly be realised, why it would involve undefined additional expense, and why, on the contrary, little benefit can be expected. On the one hand, digitization seems to be the magic word for the imperative and necessary improvement, for a desired increase in efficiency and simplification of work, but on the other hand, there is often a lack of courage to tackle such a project in one’s own operating environment and to engage in digital transformation, i.e., to take steps towards digital change. A literature survey conducted by Vial [2] mentions corporate inertia and resistance of employees as the two most relevant barriers of digital transformation. One concrete reason for this reluctance is probably the high level of expectation towards digitization on the one hand, and on the other hand, every operating environment is special, has often developed heterogeneously over the years, follows established processes, and was ultimately shaped by different people with different qualifications and viewpoints. From the authors’ experiences, for example, there are often very contradictory positions between the production and the in-house IT departments on questions of digitization, in connection with the introduction of new software systems or on interventions in the company’s own data network structures, which first have to be overcome. To ensure the success of digital transformation, the following basic advice should be observed:

  • Limitation to a clearly defined problem or area of the company;
  • Clear definition of the improvement goals to be achieved through digital transformation;
  • Identification of the processes that are to be digitized or supported by digital transformation;
  • Identification of the real and measurable benefits to be achieved through digital transformation measures;
  • Inclusion of the people who are affected by the digitization activities;
  • Digitization is a continuous learning process and uses the tools of project management and change management for realisation;
  • Digital transformation initially means additional effort and the dissolution of existing established mindsets.

The measurable benefits of digital change are often not clear at first glance. After all, at the beginning there are expenses for the project work, the procurement costs of hardware and software, and the extensive qualification of employees that have to be justified. Again, everyone will agree that digitized processes based on objective data could bring more transparency, better traceability, better process documentation, added value and perhaps even a time advantage. But what monetary advantage these improvements actually imply in operations often remains unanswered. In general, it is helpful to become aware of the situation in one’s own operating environment and to develop a realistic expectation of digitization. The question is – where do I stand in my company? There exists exhaustive literature on the assessment of the level of digital maturity, e.g. Remane, et al. 2017 [3] or Gill and Van Boskirk 2016 [4]. A simple guidance on how to assess a company’s own digital maturity level is given as follows:

  • No digitized processes;
  • Slightly, simply digitized processes;
  • Several analogue and digitized processes side by side;
  • Largely completely digital process control;
  • Continuous digital production control, process control and proactive machine condition monitoring.

A distinction must be taken here as to whether digitization is understood only as an approach to electronic information logistics, for example through the use of e-mail instead of paper, or far beyond this as a fundamental approach to datadriven process analysis for fact-based decisions. In the following, the approach of data-driven process analysis is discussed further.

2 The machine as a data source for data-driven process analysis

At the beginning, obviously besides the problem statement itself, there is always the question of data as the basis for any kind of digitalization. However, the question would be better asked after seeking the right and valuable data. Data is the treasure that needs to be mined and processed into added value:

  • Do I already have data?
  • Can I generate missing data and if so, how?
  • Do I have sufficient data?
  • Are the data already qualified enough?

Data can be used to generate information that leads to analyses and ultimately to fact-based decisions. In industrial production, it is above all operating equipment and machines that are suitable for supplying valuable data [5]. In a typical PM or advanced ceramic production, the powder presses are at the top of the value chain. Modern powder presses, like all CNC machines today, already use a lot of digital information for control, safety functions, as well as process control and general condition monitoring. Typical examples are axis positions, forces, pressures, flow rates, temperatures, functional states, counters, times, fill levels, or the electrical power consumption. By means of discrete transducers, these data can usually be easily acquired. The focus of such data is on control and process control of the equipment. The acquisition of inspection data on the press part for quality control or the acquisition of data for condition control of the operating equipment is often not available or is limited to the most important monitoring data, e.g., oil level in the hydraulic tank. Such standard data already provide a sufficient basis in many cases. In connection with further external data, e.g., data on powder and tools, the manufacturing process of a pressed part can already be sufficiently documented over the time of production, and comparisons can be made with previous productions. If, however, there is an interest in recording certain facts in the production process or for special requirements of quality assurance and machine condition monitoring beyond the possibilities of the standard configuration, then in many cases the only option is to upgrade the machine with suitable, additional measuring sensors. But retrofitting presses means additional investment, which must be set against a benefit. This is where the preliminary target definition in a digitization project comes in because it makes little sense to upgrade a press to an overly comprehensive measuring machine, either technically or in terms of cost. It is therefore important to consider well in advance which data is needed for the subsequent documentation, analysis, and evaluation steps so that added value can be generated through digitization. Many general-purpose, commercial digitization products require the existence of sufficiently expressive basic data based on discrete measurements on the equipment. The assessment of whether this basic data is sufficient for the desired objectives of the digitization project and whether the desired added value is achieved is often left to the operator, with the risk of being easily overwhelmed. In contrast, industry-specific digitization solutions from the manufacturer of the presses and equipment offer a very good alternative. The manufacturer knows his machine best, is familiar with the systems already in use, any existing interfaces and bus systems as well as the possibilities of the control system. Furthermore, he can offer unique, qualified support to the operator to define additional data acquisition points on the machine in a targeted manner and integrate them cost-effectively. The shared profound understanding and experience of the manufacturer and operator about the processes that run on the machine are of immense value here. Ultimately, this is the most effective way to translate raw data into decisionsupporting information.

3 Adding value by refining data into information

Discrete measured values from the press and additional external data together form the basis for generating further information by methods of agglomeration, filtering, and correlation. By correlating collected data, valuable information is created that can reveal new contexts or unleash interesting, potentially hidden dependencies. It must be clear that the sole existence of correlation [6] (or dependencies) within data does not necessarily lead to any useful information. It is therefore important to find links between observed outcomes and the hidden patterns behind measured values, possibly at specific points in time. Simple correlations can be illustrated well, for example, by linking data values by two characteristics in so-called scatter plots. For instance, the oil leakage of a press that typically goes with a higher power consumption can be displayed in a two-dimensional feature space. It the two characteristics are positively or negatively correlated, then the data points in the scatter plot are all located close to a straight, diagonal line. More complicated are cross-correlations with more than two parameters, which are often of interest and usually require more complex data analysis procedures. For example, there might be additional causes that can affect the power consumption. Even in the absence of oil leakage the power consumption might still be high, e.g. because the press is operating at higher pressing forces. That means that in many situations it might not be sufficient to only consider and treat two parameters in isolation. In powder pressing, important data usually do not occur simultaneously, but follow the events in the pressing cycle. Thus, even with inline systems, the weight of a pressed part or the expanded geometry can only be measured after the press cycle. However, if the assignment of the weight or the actual part height to specific data from the pressing cycle of the part is of particular interest, then the related data must first be filtered out of the continuous mass data stream and correctly assigned to the respective pressed part. This is where powerful and intelligent data pre-processing plays a key role. In the course of further data preparation, it makes sense to categorize the data and information made available in this way. With the powder press as the first singular data source, the perspectives

  • Production data
  • Quality data
  • Machine condition data have proven themselves in practice.

3.1 Production data acquisition

Production data acquisition is a largely self-explanatory procedure and is almost a standard component of machine visualizations and classic production control systems. Typical variables are the number of strokes, pieces produced, good and rejected parts, machine operating and downtimes, or the remaining production time, which are related to order data and other operating equipment data such as tool IDs or powder type batches. If appropriate measuring sensors are available, data on the energy and media consumption of the equipment provide further information which can be related to order and production data and thus enable statements to be made, for example, about the manufacturing costs per press part. With the accompanying recording of environmental conditions in the production field, additional interesting correlations can be shown, such as the tendency of powder materials to adhere to the tool and the cleaning cycles required as a result, depending on the current temperature and humidity. Only through the target-oriented data acquisition, data refinement and the simple provision of the information obtained and verified in this way are objective, personnelindependent decisions possible. In many cases where absolute measured values are not available, quality statements are based on whether a pressing process is stable or not. Once a process or operating condition is found to be sufficient for the desired quality, the goal is to keep this condition as steady and reproducible as possible. The scattering range should then only be influenced by a few disturbance variables, e.g., powder parameters. In individual cases, long-term observations can even be used to predict the influence of individual interfering variables on the quality of the pressed parts. However, if quality characteristics of a pressed part can be measured and evaluated directly, this is naturally preferred because of the higher informative value and the security gain. The quality evaluation of a process down to the pressed part level via status parameters and their stability represents an indirect procedure.

3.2 Machine condition data

Machine condition data provide a direct view of the equipment itself and are suitable for use in new digitized processes from several aspects. From the first moment of commissioning, every equipment is subject to alteration due to daily stress. The wear and tear of a piece of equipment can take place slowly and unnoticed, but it can also become immediately apparent in the case of exceptional events, e.g., in the case of an evident overload. Especially the slow wear is an oftenunderestimated cause for quality problems in production. The task of regular maintenance and recurring servicing is to reveal the process of wear, to slow it down, and to prevent an uncontrolled failure of the equipment as far as possible. The goal is to maintain the best possible machine condition over the life cycle of the equipment so that it is always reliably available for production, i.e., in a way that can be planned. The continuous recording of condition data, especially the calculation of a wear index, represents a major challenge. It is often not technically possible to use appropriate transducers for direct wear detection. Such approaches sometimes fail due to the size of transducers, installation options at the site of the occurrence, the measured value resolution or simply the costs. Not every potential wear or failure of a mechanical machine element or electronic component can be intercepted with monitored measurement technology. Instead of only measuring the direct wear on an assembly, machine element or, more generally, the alteration of the machine over its service life, continuous observation of the press cycle on the part with the associated path curves and secondary data has therefore proven its worth in practice. However, to conclude this holistic approach from the perspective of the pressed part to the machine condition requires a deep understanding of process control, control behaviour and machine properties. This means that the condition assessment of the machine and its potential, technical availability is no longer carried out on the basis of specific measured values alone but is based on data-based observation of the long-term behaviour of the pressing process. This extended condition assessment cannot usually be carried out by the operator alone due to a lack of dedicated detailed knowledge of the press, its installed components, and the control system. Here, the manufacturer of the press can actively support with its specialised knowledge by accessing the data of the machine, its collected load profiles and current snapshots and provide customer- specific consulting services (Fig. 1). Such digitalized service models relieve the operator in that the machine is under regular and online monitoring by the manufacturer and the operator benefits directly from the accumulated specialist knowledge of the press manufacturer. With a benefit as similar as in social networks, and in economic terms referred to as positive network effects [7], the individual can participate in the experience of many operators in a protected and anonymous manufacturer community.

4 From classical statistics and probability to machine learning methods

An appropriately equipped powder press in production continuously delivers discrete data in large quantities, either with each cycle, or at different times, and in different quality. For example, temperature changes are rather inertial events, while positional data from press axes are generated in large quantities and in very short time periods. Many discrete data points already carry enough inherent information to be able to document the processes during machine operation and do not require any pre-processing or refinement steps such as filtering or correlation. Typical examples for this are the recording of operating times and downtimes, or the stroke counter of a press. More advanced possibilities for the analysis and evaluation of massive amounts of data, so-called big data, are provided by the methods of statistics and increasingly by the methods known as machine learning. On the one hand, it is easy to imagine that nowadays the entire pressing cycle for each pressed part can be recorded across all axis movements, positions, speeds, torques and forces, even at high sampling rates. However, filtering and target-oriented clustering of such mass data for characteristic features that indicate a faulty pressing and thus a potentially faulty pressed part are complex and not practical to do manually with spreadsheets, if only for reasons of processing speed. Simple statistical methods such as the determination of mean values from data series, minimum or maximum values from data series (descriptive statistics), trend analysis, or boundary value analyses by using predefined envelope curves, etc. are already common methods that many modern presses come with as standard. Even the simple differential comparison of individual target values with their actual values to identify process-related scatter is not uncommon. More advanced and promising here, however, are the approaches of machine learning (e.g. [8]), in which data-based learning algorithms are used to (partially) automatically extract certain patterns and regularities from the properties of existing data sets, which in return can be used to classify and predict newly acquired data sets. Such automated pattern recognition processes based on high-resolution press cycle curves have enormous potential to draw conclusions about part quality from recognised anomalies in axis movements or force curves. Of particular interest here are small or infrequently occurring anomalies that always occur in a process that is considered inherently stable, but which are ignored when viewed as a whole. If such transparent anomalies are correlated with further data, e.g., on machine conditions such as oil temperature, power consumption, data on ambient conditions or also with the times of operator interventions during the production batch, even more comprehensive structural patterns can be generated for evaluating process and part quality over an entire batch down to the individual press part. As far as marking systems, e.g., laser matrix coding on the part, are used, the subsequent identification of a single pressed part produced with a selectively deviating press cycle is possible. By continuously collecting data at the machine with appropriate software tools, the human observer first learns through personal evaluation whether a detected anomaly or feature cluster is detrimental to part quality or responsible for part scrap due to high crack probability. However, if a generalisable statistical model can be derived from the feature patterns, the press can either sort out potentially defective parts itself without dedicated measuring equipment, e.g., a technically demanding and also expensive inspection for cracks, or in a second step even take its own actions to avoid further defective parts. In the perspective of the further use of artificial intelligence methods in powder pressing, systems are conceivable which, for example, can carry out certain start-up and set-up processes faster, more reliably and more consistently. Today, a filler axis movement is still programmed by hand and optimised with a lot of experience. In the future, why shouldn’t the press independently determine the optimum filler shoe movement in a ramp-up cycle based on learned behaviour patterns in connection with powder and tool data, and then independently adapt it to the requirements again and again during ongoing production? Data-driven digitization is becoming the motor for the future development of semi-autonomous or, later, completely autonomous powder press systems.

5 What does the practical implementation look like when you want to introduce a digitization project in your production?

Changing processes through digitization and digital tools based on data in production should always be viewed as a strategic project. As in many other strategic projects, the expectations and goals set can be implemented much more effectively and quickly with the help of external support than if you rely solely on your own forces, who are also involved in day-to-day business. Practical experience shows repeatedly that existing mindsets in production as well as in administration and IT departments must first be disrupted to open up to the digital transformation. Many project plans are initially characterised by uncertainty and great respect for the complexity of the task. Often, there is a lack of skills on the part of those affected in production and in the functional departments. For an outside observer, it is also not unusual to find that the production and IT departments of a company know far too little about each other to be able to successfully carry out a useful digitization project together and often simply do not speak the same language [2]. Either such a project then fails from the outset or, at best, they can agree on a minimal consensus that is not very efficient. There is then a considerable gap between the original ambition and the reality that is finally achieved. Negative experiences such as these ultimately prevent the courageous approach of further steps in the digital transformation and lead to frustration among all those involved. How can this be counteracted, and the right start made in a digitization project?

5.1 Management vision

At the beginning, there should always be a management vision of what you want to improve with the change through digitization. However, to ensure that this vision does not remain vague, it is necessary for decision-makers to familiarize themselves with current technologies and possible digitization designs. Meaningful process digitization is possible in every company and is not a question of size, but not every design fits everywhere and can therefore simply be copied. It is important to find the digital process design that best fits the requirements of the company. In this context, it is also interesting to recognise the so-called digital maturity of the company, so that achievable goals can be developed from the vision.

5.2 Objectives

The objectives formulate what is to be achieved through the digital change process. Digitization is never an endeavour in itself, except perhaps in terms of a marketing strategy, but must always offer a clear benefit. This expected benefit must be defined at the beginning in order to select the right process change measures and tools. We are often asked by customers what benefit this or that software, a digitization tool or the investment in the technical upgrade of machines has in production. Of course, one can then describe the benefits and the great possibilities of the software products. But at that stage, can you really credibly convey what, for example, increased process transparency offers in terms of actual, measurable added value for the customer in his particular situation? For example, what value-added benefit can be derived from the fact that a conditionmonitored machine will experience fewer unplanned breakdowns over its lifetime, or that potential damages will be reduced? A measurable evaluation would have to be based on facts that are not supposed to happen. An existing loss or production failure can be easily assessed. However, the non-occurrence of a failure is even more difficult to evaluate because one can only rely on often arbitrary assumptions and probabilities of occurrence. Consequently, it is up to the operator to define what measurable benefits are to be achieved through digital change. The digitization partner must then provide the answers to these requirements, whether its product can fulfil them and under which conditions. If an operator is not yet in a position to specify its goals and expectations for measurable benefits, it is helpful to first discuss the question of goals and benefits in one or more project discussions. These discussions can be held internally, but it has proven useful to have external support in this phase, which can analyse the current production process with a different perception, for example, and provide practiceoriented advice through special knowledge of the machine and operating equipment. In addition, external support offers a good approach to mediate between the different interest groups in the company and to reduce mutual reservations of the specialist departments towards digitization through moderation.

5.3 Qualification

Digital transformation inevitably leads to the introduction of new technologies and tools that find their way into day-to-day business, and gradually replace previously proven procedures. For the success of the digitization project, it is therefore essential to involve the affected employees from the outset and to train them in parallel with the progress of the project. It is essential to bring the new knowledge and existing experience into line with each other.

5.4 A policy of small steps

Digital transformation is a daunting task because it generally attacks tried-and-tested approaches and requires new thinking in abstract systematics. Therefore, the policy of small steps is recommended, but without losing sight of the big goal.

5.5 Time as a factor for success

Digital transformation in production is not a pill that, once dosed, automatically makes everything right and better. It may seem contradictory that digitizing processes intended to make them faster and more efficient first require time to be implemented and to mature. After all, investing in digitization efforts is subject to exactly the same criteria of economic viability as any other project. But digitization projects, in particular, are very often virgin territory in contrast to normal investment projects of capacity expansion, for instance. The time required to reorganise processes on the basis of digital transformation until measurable improvements are achieved is often underestimated. Of course, the level of success and the time a digital transformation is going to take significantly depends on an organisation’s culture, on the leadership, and on the employees’ capabilities and willingness for change [2].

6 Perspectives

The chances for improving and increasing efficiency by digital transformation of production processes are good nowadays. Powerful computer hardware, functional measurement sensors and industry-specific standard software are available. Surprisingly, however, the end-to-end datadriven digitization of industrial processes and corporate organisation is often not yet as important as one might expect. After all, private life often appears to be much more digitized than the day-to-day business of industry. The reasons for this may be a lack of identification on the part of the people involved and a lack of knowledge about digitization, but above all a lack of awareness of the real benefits and added value to be able to justify such investments. Yet digital transformation, properly introduced and lived, is more of an opportunity than a threat and paves the way for Smart Factory solutions and intelligent machines. Under the market name DORST IoT Solutions (Fig. 2), DORST TECHNOLOGIES offers a specially developed hardware and software package, which connects DORST powder presses in the production field with each other and in this particular way turns them into valuable data sources.


[1] Hype Cycle for Emerging Technologies, 2021: gartner/3-themes-surface-in-the-2021-hypecycle- for-emerging-technologies (retrieved on 6.8.2022)

[2] Vial, G.: Understanding digital transformation: A review and a research agenda. The J. of Strategic Information Systems 28 (2019) [2] 118–144

[3] Remane, G.; et al.: Digital maturity in traditional industries – an exploratory analysis. 25th European Conference on Information Systems (ECIS), Guimaraes, Portugal, 2017

[4] Gill, M.; Van Boskirk, S.: The digital maturity model 4.0 – benchmarks: Digital business transformation playbook. Forrester Report, 2016

[5] Sisinni, E.; et al.: Industrial internet of things: Challenges, opportunities, and directions. IEEE Transactions on Industrial Informatics 14 (2018) [11] 4724–4734

[6] Rodgers, J.L.; Nicewander, W.A.: Thirteen ways to look at the correlation coefficient. The Amer. Statistician 42 (1988) [1] 59–66

[7] Katz, M.L.; Shapiro, C.: Systems competition and network effects. J. of Economic Perspectives 8 (1994) [2] 93–115

[8] Mitchell, T.: Machine Learning, Boston 1997