1 Introduction

More and more aspects of our lives are consciously or unconsciously captured and stored by data. This is especially true in production. According to IBM, a modern factory generates 1 TB of data per day (IBM 2022). At the same time, data modeling methods, for example, based on artificial intelligence, are becoming increasingly powerful. The combination of both aspects makes it obvious to use data-driven modeling methods in production in order to support decision-making processes with “what-if” analyses (acatech 2021).

The applied decision support methods depend on the level of production management, which can generally be divided into two categories: short-term and long-term production management (cf. Figure 1). Short-term production management includes all operational processes on a machine and shop floor level and must especially deal with a high volatility of the environment.

Fig. 1
figure 1

Time sensitivity and uncertainty of decisions in short-term and long-term production management

Full size image

Long-term production management includes processes on the level of the whole factory as well as a worldwide production network with several locations and must mainly conquer the high uncertainty of the environment. The different expressions of volatility and uncertainty influence the decision support methods being applied in short- and long-term production management (Ivanov 2018).

To quickly compensate for disturbances within the production system, time sensitiveness of decisions is decisive, which is why the decision support methods in short-term production management aim to significantly increase decision and implementation speed. To deal with the high level of uncertainty within the production system, decision support methods in long-term production management aim to maximize decision quality.

This chapter presents data-based methods to support decision-making in production management while dealing with time sensitiveness and uncertainty. These methods are developed on the basis of six use cases, as shown in Fig. 1. Thereby, decision support is realized with so-called applications (apps) focusing on the information need of decision-makers in production. The following sections present these applications. Section 2 focuses on short-term production and thus on increasing decision and implementation speed. Section 3 focuses on long-term production and how to increase decision quality.

2 Increasing Decision and Implementation Speed in Short-Term Production Management

The volatility of today’s markets is constantly rising due to the rapid emergence of new and innovative competitors, changing government policies, as well as unknown market acceptance – to only mention some factors. Therefore, production is more and more characterized by shorter product lifecycles, increased individualization, and disruptive technological changes (Schlegel et al. 2018). The tasks of short-term production management thereby include, i.e., occupancy planning, resource monitoring, quality control, and order scheduling. Thus, short-term production management must secure that change requests or disruptions on the shop floor (e.g., machine failure) are compensated via appropriate control loops. A quick response to changing circumstances and requirements is crucial to achieving corporate goals. Consequently, the ability of short-term production management to carry out process adjustments efficiently is of particular importance (Petschow et al. 2014).

A key requirement for efficient process adjustments is the reduction of the overall latency between the occurrence of an event and the implementation of the derived corrective measures. The lower the overall latency, the higher the value creation of production, as shown in Fig. 2 (Zur Muehlen and Shapiro 2015; Kiesel 2022). Overall latency is composed of four delay times: data transfer latency, data analysis latency, decision-making latency, and implementation latency. Data transfer latency refers to the timespan between the occurrence of the event and the moment when data is available for analysis. Data analysis latency is the time of initiating the analysis, packaging its results, and delivering them to the appropriate system. Decision-making latency is the period the system records this information and takes a decision. Implementation latency describes the time between the decision and the execution of the corresponding measure (Hackathorn 2002; Kemper et al. 2010; Sejdic 2019). Thus, to pursue the goal of overall latency reduction, the four presented sub-latencies must be reduced.

Fig. 2
figure 2

Potential efficiency enhancement by reducing latencies in short-term production management

Full size image

In order to compensate disturbances within the production system as well as to implement changes with the minimum latency possible, the concept of self-learning production systems is further developed within the Internet of Production (IoP) (Brecher et al. 2017). Within this concept, the availability of real-time shop floor data and machine learning algorithms enables process models to learn from historic process data. This in turn enables the prediction of product quality as well as the adaption of production processes accordingly (Lee et al. 2014). To further develop the concept of self-learning production systems, the IoP pursues three main objectives. First, IoP develops data-driven methods (e.g., in the field of process mining) that enable self-learning production systems to learn from historic process data and events and make autonomous decisions (van der Aalst et al. 2020). Second, IoP drastically minimizes data and analysis latencies through the integration of continuous cross-domain data access and the development and combination of diagnostic, predictive, and prescriptive analytics models. Third, IoP reduces decision and implementation latencies by means of an appropriate collaboration of autonomous processes and model-based decision support as well as the implementation of suitable measures in the production system.

With these three objectives, IoP increases productivity by reducing the impact of volatile environments on a steady production system. IoP furthermore masters quick change requests by decreasing the period of time which is required to bring the production system back into a steady state after process adjustments.

To better understand both the objectives of IoP and their impact on short-term production management, three applications are presented in the following. Table 1 shows their overall goal and their contribution regarding IoP’s purpose.

Table 1 Exemplary applications for short-term production management and their purpose within the IoP
Full size table

2.1 Predictive Quality (Quality Control Loops)

Manufacturing processes have become significantly more complex in the past years due to the ongoing digitalization and interconnection of systems. Early defect detection in interlinked production steps offers the chance to reject affected parts at an early stage of the production process so that costs and efforts for dispensable further processing can be avoided.

Within the IoP, early defect detection is realized by predictive defect models. It enables companies to identify problems in process and product quality at an early stage of production by providing the employees with the information they need to enable data-driven decision support via the application. The models consist of machine and inspection data being interlinked with the process variables that have an influence on defect occurrence.

To develop the predictive defect models coping with the requirements of time sensitiveness of decisions, a flexible process-independent meta-model for production data is developed. This model is the basis for the development of data-driven methods. Based on the context provided by the meta-model, automated machine learning (AutoML) methods for pre-processing and analysis of production data are developed. By automating the ML pipeline, AutoML significantly reduces data and analysis latencies (Schmitt et al. 2021).

2.2 Short-Term Production Planning and Control

Production planning and control (PPC) is a highly complex task in job shop production. Companies often struggle an economically beneficial operation strategy in the polylemma of short lead times, low working capital, high utilization, and high adherence to delivery dates. To tackle this polylemma, IoP uses feedback data for an optimization of PPC tasks (Schuh et al. 2020).

Therefore, IoP focuses on the development of a reinforcement learning agent that uses realistic simulation models of a job shop production for learning and optimizing the task of order release. The simulation model allows an instant reaction to production disturbances via order rescheduling, rerouting, and changing of dispatching rules. Thus, considering current production goals, production utilizations are aligned and optimized. The simulation model is generalized over different types of production (e.g., mass customization, craft production, batch production) using transfer learning.

Besides the development of data-driven methods using reinforcement and transfer learning, the app especially enables a reduction of decision and implementation latencies by an autonomous decision preparation based on digital shadows.

2.3 Parameter Prediction (Production System Configuration)

In several industries, e.g., the textile industry and plastics production, to date, manual process and machine adjustments are the norms rather than the exception. Thereby, the correct setting of the machine depends on many different parameters and often requires knowledge of an experienced engineer (Müller et al. 2023). To become less dependent on expert knowledge of these engineers and shorten the machine setting duration, the goal of this app is to predict machine settings based on specific process data, especially quality parameters.

To do so, a holistic machine learning model is created within the IoP. It is based on reverse neural networks (RNN). Based on historical or synthetically generated data, the RNN identifies correlations between process parameters and part quality and then calculates process parameters that can be used to produce the desired component with the required properties. This facilitates the induction of new employees in industries dependent on expert knowledge and at the same time objectifies existing domain knowledge. Furthermore, a prioritization between the target parameters of a production system is provided. If, for example, energy consumption is a priority and fast processing is negligible, these specifications can be implemented as an optimization problem and adapted recommendations issued (Müller et al. 2022). Besides the development of predictive models, this app increases the decision as well as implementation speed in production, as parameters are objective and no further decision is required by the machine operator.

3 Decision Quality Enhancement in Long-Term Production Management

Long-term production management considers the entire supply chain network and the internal production network. The main goals of long-term production management are cost reduction, flexible production structures, resilience, and sustainability (Lanza et al. 2019). Long-term production management thereby determines the future production structure and consequently has a high impact on the long-term competitiveness. Since long-term decisions often require substantial resources, may be irreversible, and define an organization’s direction for years to come, the decision quality is of particular importance. Thus, over the last years, data-based decision support systems have been identified as an opportunity to support decision-making processes (Tiwari et al. 2018).

However, achieving a high decision quality through data-based decision support is very difficult, since decisions in long-term production management occur uniquely and infrequent and are subject to a high degree of uncertainty. This uncertainty often results from different and unreliable internal and information sources, such as sales forecasts or market demands (Lanza et al. 2019). In addition, new production platform models require a new degree of openness between market players, which lead to shifting property rights and decision responsibilities and hence a new dimension of uncertainty. While increasing openness can increase value creation, it might be a risk to control value capture (Schuh et al. 2018).

Thus, the main lever to increase decision quality in long-term production management is the reduction of uncertainty in decision situations. Within the IoP, we thereby distinguish between five types of uncertainties occurring in between the need for a decision and its final implementation, as summarized in Fig. 3. Action uncertainty describes whether the event requiring a decision will occur at all. Scope uncertainty is the uncertainty about the environment the decision will affect (Welsh and Sawyer 2010). Data quality uncertainty includes the uncertainty of trustworthiness and reliability of the available internal and external data. Prediction uncertainty refers to the variability in prediction due to plausible alternative input values (Tavazza et al. 2021). Decision uncertainty refers to the variability in decision implementation due to different decision alternatives. Thus, to increase decision quality, these sub-uncertainties must be decreased.

Fig. 3
figure 3

Potential of decision quality enhancement in long-term production management

Full size image

Therefore, IoP’s vision is to improve decision quality by supporting the decision-maker in the proactive design and improvement of production structures in uncertain business environments through intelligent decision methods and the respective autonomous algorithms. To realize this vision, IoP pursues three main objectives. First, adjustment requirements in production structures are proactively identified by continuously monitoring relevant events in production, product development, and usage. Second, intelligent methods for an autonomous decision preparation are developed based on digital shadows. Third, IoP analyzes how the decision-maker can be supported by a comprehensive suggestion of alternative courses of action and the assessment of their impact on strategic targets.

With these objectives, IoP delivers higher transparency and trust over decision needs, influencing factors and uncertainties as well as the impact of domains like product development and usage. IoP thus enables to continuously monitor and adapt the long-term targets of production. Furthermore, IoP allows a new way of strategic decision-making by autonomous decision preparation, analysis, and support. This enables decision-makers to focus on the value-adding part of long-term decisions in designing future production structures.

To better understand both the objectives of IoP and their impact on long-term production management, three applications are presented in the following. Table 2 shows their overall goal and their contribution regarding IoP’s purpose.

Table 2 Exemplary applications for long-term production management and their purpose within the IoP
Full size table

3.1 Proactive Factory Planning

Factory planning projects often fail to comply with time and budget restrictions and thus expose enterprises to a variety of risks. Especially the planning of greenfield factories with an almost infinite solution space – thus many uncertainties – entails the risk of wrong decisions during the planning process. Studies identified information management to lie at the root of this problem, as the information in the planning process are often interconnected and must therefore be managed suitable methods to ameliorate factory planning outcomes. Otherwise, the interconnection leads to even higher uncertainties, which affects the transparency of decision and probably reduces decision quality (Burggräf et al. 2021; Herrmann et al. 2020).

To reduce these risks, risk management must be part of the factory planning process. Risk management aims to identify, assess, and prioritize individual risks of information so that appropriate decisions and actions can be taken. However, standardized methods are not implemented in factory planning yet (Burggräf et al. 2021). Therefore, IoP develops a new risk management method for factory planning. Especially the scope uncertainty (cf. Figure 3) shall be reduced by this methodology.

The risk management approach bases on fuzzy logic methods. Fuzzy methods allow to model and calculate decision-making deficits and uncertainty of different stakeholders (Bellman and Zadeh 1970). Thus, in the factory planning process, fuzzy logic approaches can be used to make information uncertainties measurable.

Where otherwise only subjective estimates can be obtained, fuzzy logic can contribute to capture information for factory planning scenarios transparently, objectively, and quantitatively (Burggräf et al. 2021).

This objective quantification reduces the uncertainties and thus increases the decision quality in factory planning. Furthermore, fuzzy logic-based risk assessment enables an autonomous decision preparation. It also allows a proactive planning process as risky planning steps are known in early stages of the planning process and can thus be prevented from the beginning.

3.2 Supply Chain Cockpit – Master Data Quality Improvement

As past disruptions of the supply chain, e.g., COVID-19 or the blockage of the Suez Canal, have demonstrated, their impact on the procurement side is highly critical. Procurement is responsible for organizing and ensuring the supply of external material and parts that are required for internal processes. Thus, forward-looking procurement planning which is prepared for disruptions is key for production companies (Linnartz et al. 2022).

Thereby, procurement planning is often realized within an ERP system and requires data from various existing and potential suppliers. Data quality thereby affects decision quality significantly. Due to its various sources, data of procurement often lacks quality (Ge and Helfert 2013). Therefore, the IoP developed an app to practically identify, prioritize, and take measures against poor data quality.

This app thereby executes two main tasks: First, it identifies critical data quality problems within the master data of the ERP. Second, based on the identified lacks, the app derives recommended actions and different alternatives to improve data quality. This way, the app enhances trustworthiness and transparency of the data on which procurement decisions rely on. It furthermore proactively identifies adaptation needs in master data quality and recommends alternative courses of action toward improvement. This way, it contributes to an enhancement of the decision quality within long-term production management.

3.3 Footprint Design (Production Network Planning)

Sustainability of global production networks is critical. While still being efficient and profitable, production companies must secure sustainability of its network (Alexander 2020). Therefore, global production networks should be continuously evaluated and improved regarding their sustainability characteristics. To do so, IoP develops the Footprint Design App, which is designed for production network planners to proactively identify adaptation needs in network design regarding sustainability before the design decision of a network configuration.

The core of this app is a data model to combining production network elements with sustainability characteristics and attributes. This model is fed with data from existing production networks, internal company data, and data from external sources such as LCA databases and transport information. This data is then related to the sustainable characteristics and attributes. This way, the app allows an optimization of global production network footprint.

Decision quality is thereby enhanced in several ways. First, by continuously updating the database and reconfiguring the network, data quality increases, and overall uncertainty decreases. Second, by proactively identifying adjustment requirements as well as suggesting alternatives of network design, the scope of the decision is narrowed down for the production network planner, which again reduces uncertainties.

4 Conclusion

Data-driven modeling methods in production to support decision-making processes with “what-if” analyses are of high importance for future production management. Therefore, the IoP develops apps for decision-makers in production to support the decision-making process.

In short-term production management, they support reduction of latencies in the decision process and therefore a faster implementation of decisions, as it was illustrated by the three apps “Predictive Quality,” “Short-Term Production Planning and Control,” and “Parameter Prediction.” Thereby, it was shown that data-driven decision support increases productivity as the impact of volatile environments is reduced. Additionally, quick change requests can be mastered in a much shorter time.

In long-term production management, data-based decision support reduces uncertainties and thus increases decision quality, as it was illustrated by the three apps “Proactive Factory Planning,” “Supply Chain Cockpit – Master Data Quality Improvement,” and “Footprint Design.” Here, decision support increases transparency and trust of decisions. It furthermore allows new ways of strategic decision-making by autonomous decision preparation, analysis, and support.

This way, IoP secures that business goals are reached and market needs are met by improving the decision quality and implementation speed in long-term and short-term production management.