Monetization of manufacturing data
Components, Challenges and Solutions for Industrial Data Marketplaces in the Manufacturing Industry
Co-Authors: T. Bergs, C. Gülpen, W. Maaß, Johannes Mayer, H. Musa, Philipp Niemietz, F. Piller, A. Rohnfelder, M. Schaltegger, S. Seutter, J. Starke, E. Szych, Martin Unterberg
Dieser Artikel ist auch auf Deutsch verfügbar.
Prologue
In 2018, the vision of a data economy for manufacturing technology (a must-read) was outlined together with the WZL of the RWTH Aachen University within the Industrial IOTA Lab Aachen – founded by senseering GmbH and the WZL. This contribution continues the vision and focuses on possible monetization strategies to formulate incentives in economy and society.
Preamble
This article has officially been published on:
If you want to cite this article, please use:
Trauth, D.; Bergs, T.; Gülpen, C.; Maaß,W.; Mayer, J.; Musa, H.; Niemietz, P.; Rohnfelder, A.; Schaltegger, M.; Seutter, S.; Starke, J.; Szych, E.; Unterberg, M.: Monetarisierung von Fertigungsdaten. In: Internet of Production — Turning Data into Value: Statusberichte aus der Produktionstechnik 2020 (Herausgeber: Bergs, T.; Brecher, C.; Schmitt, R.; Schuh, G.). Aachen: Fraunhofer IPT, 2020, Seiten 377.
Preface
This article was prepared based on the results of the expert group of the Aachen Machine Tool Colloquium AWK 2021. The following experts were involved:
Dr.-Ing. Dipl.-Wirt.Ing. Daniel Trauth (senseering GmbH), Prof. Dr.-Ing. Thomas Bergs (WZL der RWTH Aachen), Christian Gülpen (TIM der RWTH Aachen), Prof. Dr.-Ing. Wolfgang Maaß (DFKI GmbH), Johannes Mayer (WZL der RWTH Aachen), Heiko Musa (BMW Group), Philipp Niemietz (WZL der RWTH Aachen), Frank Piller (TIM der RWTH Aachen), Andreas Rohnfelder (Fujitsu Deutschland GmbH), Markus Schaltegger (Feintool AG), Sebastian Seutter (Microsoft Deutschland GmbH), Joachim Starke (BMW Group), Elmar Szych (Dell EMC GmbH), Martin Unterberg (WZL der RWTH Aachen)
Table of Contents
- Introduction
- Data Monetization
- Potentials for Industrial Manufacturing Technologies
- Components, Challenges and Solutions
- Scenarios for Manufacturing Technologies
- Conclusion
References
Abstract
Monetization of manufacturing data enables companies in the manufacturing industry to increase their productivity and sustainability. In addition to this optimization of existing business models, monetization of manufacturing data can simultaneously open up new income streams for companies through new digital and progressive business models. Advances in the field of Artificial Intelligence and data processing allow insights to be gained from manufacturing data in a targeted manner and exchanged within the framework of data alliances. In this article different steps of data exploitation from knowledge generation to monetization of this data are conceived, technical basics are presented and resulting potentials, obstacles and their solutions are discussed and illustrated by means of practice-oriented use cases. Finally, recommendations for companies in the manufacturing industry are derived and a potential monetization strategy for manufacturing data is designed.
1. Introduction
Additional to the internal use of manufacturing data, the cross-company exchange of manufacturing data also holds a high potential for optimization and innovation. With the help of suitable artificial intelligence (AI) models, previously hidden causalities and correlations can be identified and made usable beyond company boundaries. An indirect cross-company exchange of corporate know-how was observed in the 1980s in the course of the rise of the computer industry in today’s Silicon Valley, USA. The transfer of employees and their knowledge to competing companies had a disproportionate effect on the growth and innovation of the entire industry, including the companies that were abandoned. This effect generally referred to as knowledge spillover [1] is based on research by economists Alfred Marshall, Kenneth Arrow and Paul Romer (also known as the Marshall-Arrow-Romer (MAR) spillover effect [2]). According to the MAR spillover effect, the regional proximity of companies within a common industry influences whether and how knowledge moves between companies to facilitate innovation and growth. It was shown that the effect was more intense for companies in a similar industry with smaller geographical distance. The exchange of ideas and knowledge about new products and new ways of producing industrial goods largely takes place between employees of different companies in the same industry (see Figure 1). The opportunity to exchange innovative ideas is a key factor to new product development and improved manufacturing methods. In the recent past, the successful and widespread occurrence of the knowledge spillover effect was repeated in the regional cluster of the movie industry around Los Angeles, USA, and within the social media industry around Facebook, YouTube and Twitter [3]. In the case of complex company processes and products, employee migration alone does not lead to knowledge spillover effects [4]. Highly specialized manufacturing sequences of complex products require the interaction and coordination of many individual employee competencies. Therefore, individual employees cannot transfer this know-how to other companies for innovation purposes alone. Besides explicit knowledge of employees, company know-how is also available through implicit knowledge encoded in raw data. Raw data is usually not valuable without appropriate pre-processing and decoding.
The networking of manufacturing companies in the sense of an Internet of Production (IoP) enables a cross-domain cooperation between development, manufacturing and usage by processing and decoding raw data. Depending on the application, the raw data from these domains only becomes usable knowledge through timely availability, good organization, context, and analysis with suitable models. Data-driven knowledge spillover effects are such effects that arise from the exchange of knowledge gained from raw data within or across companies. These effects can be exploited systematically by forming data alliances even for complex processes and products by using artificial intelligence (AI) to create innovations and optimize existing processes (see Figure 1). In this context, companies must have a sufficiently interoperable degree of networking. Data models and data management systems must be networked within a data alliance so that usable knowledge is created from the provision of raw data and the subsequent aggregation and transformation with the help of suitable models [5]. The potential of using data-driven knowledge spillover effects within the framework of data alliances is diametrically opposed to the scepticism of companies with regard to data sovereignty, data privacy and, in particular, uncertainty about the value of their own data.
In this paper, different stages of data exploitation from knowledge generation to the monetization of this data are conceived, technical basics are presented and the potentials, obstacles, and their solutions are discussed and illustrated by means of practice-oriented use cases.
2. Data Monetization
In industries such as telecommunication and advertising, process, product, and usage data are used for the purpose of product and process optimization, which is why data pricing as an economic good already functions successfully in these industries [6]. This is particularly due to the fact that the availability and use of data in these industries has long been part of their product range. This enables both the usage and the exchange of data as an economic good and thus data monetization.
The term data monetization does not only include the commonly used meaning of a direct exchange of incentives for a digital date. Rather, it makes sense to extend the term including the exploitation of a date and the knowledge it contains to generate monetary added value. In the environment of manufacturing technology, this monetary added value can be an increase in processing time, quality and a reduction in costs. The monetization of manufacturing data thus describes any process that creates a measurable economic advantage by using (aggregated and transformed) manufacturing data to exploit data-driven knowledge spillover effects. Two basic categories can be distinguished.
The internal monetization of manufacturing data describes a class of strategies and processes within a company in the manufacturing industry, which optimize company and manufacturing processes or products and generate economic added value through vertical process analyses with the help of knowledge from own or purchased manufacturing data. Within the company, economic added value can be realized most quickly, since the demands on data security, data sovereignty and data protection are generally lower than with external data monetization, as sensitive data does not have to leave the company network. The added value of internal data monetization is shown in Figure 2.
With little effort, it is possible to gain new insights into manufacturing or business processes, reduce decision latencies, optimize products and minimize risks. The effects of internal data monetization accordingly aim at increasing the profit margin of existing business processes and models.
The external monetization of manufacturing data is a class of strategies and processes that provide internal manufacturing data and insights from data through appropriate analytical models via horizontal analyses along the value chain and across company boundaries, based on digital business models. Meeting data security and protection requirements is a necessary prerequisite for such a monetization strategy, because internal and possibly sensitive company data is exchanged across company boundaries. Implementation promises to open up new income streams and gain new market shares through digital, innovative and progressive business models and digital incentives (see Figure 3). The value creation of the traditional mechanical engineering industry expands in the course of an external data monetization, since in addition to physical products, digital products can be sold to collaborators and interested parties in the form of data and services.
While internal data monetization is already possible and used today in a wide variety of manufacturing processes [7], external data monetization is not yet established in manufacturing technology. One of the main reasons for this is the great scepticism of many companies to pass their data on to third parties, while being unaware of how much existence-securing know-how is represented in the data and what value their own data has. The risk of selling potentially very valuable data too cheaply is too high for many companies. This makes it more difficult to price and therefore to trade company data openly. In most cases, it is impossible for companies to explore the potential of their own data sets by using AI-based methods, which is due to the lack of their own personal AI expertise. Thus, a relationship of dependence with external service providers exists, which in turn creates obstacles in terms of data privacy and data sovereignty.
A possible solution is the use of AI-based services that make predictions based on anonymized raw data. By means of a suitable participation model, the data producers can participate in the findings and innovations resulting from their data. The value of a date is thus directly linked to the quality and availability of digital AI-based services.
3. Potentials for Industrial Manufacturing Technologies
The possible applications of the monetization of manufacturing data are manifold and promise an added economic value of more than $100 billion (see Figure 4) [8].
3.1 Plant optimization
Data on unforeseeable and unexpected machine failures is rare. However, training data-driven models to make robust predictions requires information about such unexpected events. Data of the same machine type from different machine operators enables the improvement of maintenance measures in the sense of predictive maintenance through data-driven prediction models and thus increased machine uptime and product quality [9] through data-driven knowledge spillover effects. The availability of data that can be used for plant optimization requires a minimum of sensor-based measuring equipment. Interoperability of this measuring equipment is important to ensure reliable data acquisition. Time series, which monitor the state of tool wear in the form of an acoustic signal, for example, may have to be enriched with event data from the plant’s manufacturing data acquisition in order to allow reliable conclusions about the process. Transfer learning through external data monetization and thus the purchase of data from other producers with the same or similar plants data-driven models can be enriched and refined further [10].
3.2 Product and process monitoring
The exchange of data containing product- or process-relevant information, e.g. the location of products or the status of processes, enables an immediate response to disturbances and errors in supply and value chains [8]. By knowing the current geo-position of goods or the inventory of a customer and his state of process, inventories of specific product categories and materials can be minimized and maximized as needed, also enabling a more precise planification and clocking of the processes. The difference to conventional RFID tracking is the end-to-end visibility, which is guaranteed by direct data exchange and external data monetization [11]. Errors during the delivery of parts can be compensated either by safety stocks or automatic switching to other suppliers, thus increasing the resilience of companies in the manufacturing industry. The digital representation of the process chain serves as evidence for the clarification of liability issues and predictive maintenance.
In addition, product and process monitoring enable certification-relevant proof of compliance with selected parameters, such as temperature or a maximum level of vibration during manufacturing or transport. Direct access to the relevant data, for example, also enables changing suppliers or avoiding waste.
3.3 Exchange of product characteristics
Product properties such as the exact component geometry or the material properties of a semi-finished product can be represented with the help of a digital twin. Furthermore, the exchange of a digital twin makes it possible to individually synchronize and optimize manufacturing processes. Continuous quality controls on receipt of goods become obsolete due to traceability [12]. The digital twin expands the database of a CAD model by the actual dimensions and qualities of a component as a result of the various manufacturing steps, thus enabling data-based auditing or certification. The finest possible representation of a product by a digital twin requires intensive cooperation between the most diverse partners in the value chain, but also holds the potential for both supplier and producer to optimize their operative business by means of external and internal data monetization.
3.4 Verification of provenance
Proof of provenance of selected goods is increasingly being focused by many customers. Each manufacturing step is ultimately dependent on the products delivered. A producer must give his suppliers the confidence that their products meet the agreed quality requirements, for example. Otherwise the process design would not be correct, and the end product would be defective. If the producer could acquire data in the sense of an external data monetization that depicts the product life cycle, he would have a secure base of evidence in the event of a defect. Furthermore, especially with regard to expected regulations and social changes, a verification of the provenance of raw materials, semi-finished products or products that guarantee a certain CO2 footprint is conceivable in the future. Companies could thus be given incentives to leave the smallest possible CO2 footprint.
4. Components, Challenges and Solutions
Companies have to meet certain requirements in their data acquisition and data processing strategy in order to monetize manufacturing data. Manufacturing data is raw, unprocessed and has no context. Therefore, it has no value per se unless it can be assigned to a context that gives it a meaning. Accordingly, analytical models cannot transform data into information without a context. For example, the prediction of a quality characteristic of a component has no added value if the context of the analysis, such as the considered period, is unclear. At the same time, it is critical for companies to know which information can be put into context and which not, because this allows conclusions to be drawn about critical employee key figures. This results in various components and challenges for both internal and external data monetization strategies, which are analyzed in more detail below. Furthermore, solutions developed for these challenges in the form of data alliance architectures, enabler technologies and possible participation models are presented.
4.1 Components of data monetization
Data quality is composed of the size of the available database with regard to the problem to be solved, the context in which the data is related to other data sets, and the degree of data cleansing. Data cleansing is the process of recognizing and correcting (or removing) incomplete or inaccurate data points from a data set, a table, or a database. It refers to the identification of incomplete, incorrect, inaccurate, or irrelevant parts of the data and the subsequent replacement, modification, or deletion of contaminated or coarse data [13]. After cleansing, a data set should be complete and consistent with similar data sets in the system.
In order to be able to exploit the potential of a database, companies must carry out data acquisition and data preparation using their own resources and means, since the individuality of the solutions and interfaces used in manufacturing necessarily requires individual solutions. It is the responsibility of the participants of a data alliance to define a uniform format of certain data types, which is then implemented individually by the participants.
In model building with AI, for example, statistical learning procedures are trained with the help of suitable adjusted data sets in order to identify causalities and correlations. In the manufacturing industry, additionally to event data (e.g. machine status and process status), univariate and multivariate time series are common data types. Besides traditional machine learning methods in the form of regressors and classifiers, such as Convolutional Neural Networks [14] or Support Vector Machines [15] for the analysis of time series, hybrid approaches with Fuzzy Logic [16] and Process Mining [17] are also suitable for usage in the manufacturing industry, e.g. in production planning.
Smart services are understood to be the highest level of digital infrastructures that analytically condense data of cyber-physical systems and extract knowledge that is important for decision-making. Smart Services are developed to make these findings available in a context-related, demand-oriented and value-added manner via platforms and to make them accessible to non-specialist personnel [18]. This broad usability enables innovations and increases in value of existing and new technologies and products (see Figure 5).
These basic components used for the refinement of enterprise data are necessary to make the knowledge implicitly contained in data usable for the use cases in the companies. Figure 6 shows the importance of these basic steps schematically. In order to use AI profitably, the basic requirements in data acquisition and processing must be fulfilled and a minimum of research and development (R&D) capacity must be invested in the development of data quality (database, context, and data cleansing). The risk for R&D investments in the field of AI can then be outsourced by using services of external experts.
Additional components relate primarily to data security and privacy, which includes the sensitive information contained in data about employee effectiveness, machine utilization or technological details. By making data anonymous and removing critical meta-information, these challenges can already be mitigated and circumvented.
Other factors with an impact on the value of data can be summarized as follows:
- The timeliness of data is essential in many use cases. If, for example, machine data is to be evaluated for maintenance purposes, it is essential that the data is up to date. If this machine data is only available after several days, it can still be incorporated into data-driven models, but it no longer provides information about the current status of a machine in the sense of real-time monitoring of a machine park.
- A date that potentially has many interested parties can be sold or exchanged several times. However, the value of a date can also be high due to the exclusivity of the ownership and can lose its value significantly to completely through multiple sales. For example, if a company has exclusive information about when certain machine components will fail, the required services can be prepared and made available directly to the customer without any competitive pressure. However, if competitors also possess the respective information, there is no competitive advantage by knowing the information.
- Only by developing AI models and testing them in suitable application scenarios the economic potential of certain data can be determined. If very precise prediction models are developed from certain data sets, for example for the wear of tool components, the value of the data can also increase subsequently.
- In the development of AI models, data sets with anomalies are particularly valuable, because they are often rare but indispensable for the performance of AI models. If data sets are proven to contain a high number of anomalies, or if anomalies could be identified beforehand and offered for sale, the valuation of this data can be above average.
- The extent of the context of the data with other (possibly external) sources is critical to the potential value of a date. Data can potentially be evaluated with the help of other data sources and thus allow for more far-reaching conclusions than isolated data sources.
Monetization of manufacturing data has not yet been fully implemented in industrial manufacturing technology. The reasons for this are manifold and will be explained in the following.
4.2 The failure of monetization of manufacturing data
The lack of clarity about the value of data leads to a high degree of uncertainty for companies in ex ante investments [19]. In the manufacturing industry there are often old manufacturing machines in operation that still perform their original functions but do not have the possibility to collect data. In industrial practice, these machines are usually not retrofitted with comprehensive data acquisition and processing capabilities in the form of high-frequency sensor technology and edge devices if the added value of this equipment is unclear. If in individual cases machines are retrofitted because a specific issue is being addressed, most of the time isolated solutions are created without interoperability and based on different, possibly incompatible protocols. With a lack of or insufficient acquisition of manufacturing data, it is then again not possible to train precise data-driven models that generate real added value from the raw data.
However, in companies where sufficient manufacturing data is successfully collected, only relatively few of the collected manufacturing data is used for analysis purposes (see Figure 7) [20]. This is, among other things, due to a lack of skilled workers or smart services in the field of data science. According to statistics, there are only about 300,000 AI developers worldwide, while the demand for these developers is in the millions [21]. A reliable and interoperable offer of AI-based smart services for manufacturing does not yet exist. Without professional data preparation and analysis, information cannot be extracted from collected manufacturing data.
Furthermore, companies often have concerns about the disclosure of data to third parties. A study by Bitkom e.V. and KPMG Intl. states that 74% of the companies surveyed have concerns about the disclosure of data to third parties [22]. A frequently discussed aspect is the question of the trustee, who guarantees the security and auditability of the use and purchase processes of data. Centralistic entities, which collect and manage large amounts of data from all participants centrally in one platform, are viewed with scepticism, since access to a large number of different company data is made possible via a central platform. One solution to this problem is a kind of decentralized data marketplace for exchanging and trading data. Ensuring data sovereignty and integrity requires novel technologies in order to ensure trust in the context of a data alliance between different participants. Data sovereignty is the freedom to decide when which data can or should be traded, exchanged or used. Data integrity, on the other hand, ensures that transferred data cannot be changed (see Figure 8).
A further essential question in company-internal exploitation strategies and cooperations or communities of interest between companies concerns the design of a monetarization of manufacturing data. On the one hand, incentives should be created for internal entities to exchange and use data within the company. On the other hand, data-producing and model-using companies as well as model-developing companies and platform operators, who assume the trustee function within data marketplaces, must share the added value of data trading equally. As long as the participation of the various participants in a data-based value chain is not defined, there is no incentive for data-producing companies to share data and take a potential risk, for example.
Challenges and solutions
In summary, the following four challenges for manufacturing companies will be identified:
- Lack of a corporate strategy for data acquisition
- Lack of AI competence and AI experts as well as a high financial risk when piloting AI applications due to a high ex ante investment
- Uncertainty about the potential added value (knowledge) of the data and lack of a monetary valuation basis and monetization strategy
- Lack of trust in platforms, third parties or project partners regarding data sovereignty and integrity
In many companies, these unsolved challenges lead to data silos in which data is merely collected and stored. These data silos are inaccessible to external companies and usually not yet integrated into value-adding processes within the company. These data silos are usually not designed for interoperability and are therefore unsuitable for cross-company exchange for analysis purposes. The following solutions need to be developed for a data exchange via a data marketplace and the use of innovative potential of data-driven knowledge spillover effects:
- Design of a decentralized infrastructure that collects and processes data from different machines, makes the data available and is interoperable within a company
- Development of a data privacy system that protects employees, company secrets and know-how by deliberately removing or omitting certain contextual details, e.g. time
- Design of a data marketplace that ensures data integrity, audits data-related purchase and usage processes, and implements decentralized data storage and access concepts so that no participant or provider of the data marketplace can view all data
- Develop internal and external data monetization strategies that allow companies to sacrifice risk, cost, and capacity for potentially useless data, and to outsource while sharing the success
- Development of pricing models for the use of external data sources, so that incentives are created to participate in a data marketplace and offer data
The development of these solutions, within the framework of which manufacturing data can be exchanged between companies and AI service providers securely and with benefits for all parties involved, requires the use of secure and decentralized architectures.
4.3 Architecture of a data marketplace
A data marketplace for trading and exchanging data is characterized by a digital platform that enables trading of raw data, processed data, data-based models, and data-centric services (such as visualizations). The function of the trustee can be described as an intermediary between the individual participants in the network. The function of an intermediary between manufacturing companies and AI experts or service providers for data products is particularly noteworthy (see Figure 9).
In addition to acting as a trustee and providing the infrastructure, this intermediary must determine functions for determining data quality, provenance, and the degree of refinement. Furthermore, it must guarantee the provision of a multitude of interfaces for the integration of different data sources within the data marketplace. Companies interested in participating in a data marketplace are faced with the central question of the data sovereignty of traded data. Classic centralistic models, which are managed by a single entity, are met with rejection. As soon as data is offered, it must leave the corporate network and be made available to the platform provider. This potentially allows the platform provider to access the data and use it for its own analysis purposes without the consent of the data provider. Similar phenomena are known in the Internet industry, for example, with Facebook and Google, where personal data is not only stored centrally, but also often analyzed and resold without the knowledge or consent of the users. A solution to the problem of a centralized approach to data management is offered by the combination of decentralized edge-based systems for data storage and distributed ledger-based systems for decentralized data management.
4.3.1 Data storage (Edge-based Systems)
Edge-based data storage and management approaches are essential to keep acquired data local to the corporate network. Sensitive data only leaves the internal corporate network when the company explicitly agrees. The guarantee for data sovereignty is therefore provided by the decentralized nature of the storage system. Only when trading is successful the data is passed to the buyer and leaves the network of the data generator. The data marketplace only has data descriptions (metadata) and has no access to raw data at any time. Accordingly, only previously defined meta-information describing the data set is stored centrally in a cloud architecture to enable platform participants to search for suitable data sets. The local storage of the information enables companies to independently ensure the security of the data and to reduce the network load, since only data that is explicitly requested is exchanged across network boundaries. The provision, maintenance and further development of the edge-based storage system is carried out by the platform provider.
4.3.2 Administrative sovereignty (Distributed Ledger Technology)
Edge-based decentralized data storage is already in productive use in many areas, but most of the available solutions are managed centrally by one entity. These traditional solutions do not allow the responsibility for storage and access resources to be distributed among each network participant and are instead dependent on one entity. The class of Distributed Ledger Technologies (DLT, such as Blockchain), on the other hand, as a geographically distributed and practically forgery-proof database, enables the cross-company trading of data as assets between all market players (manufacturing companies, suppliers and data scientist) to be transparent, tamper-proof and traceable.
In the context of a monetization of manufacturing data, the DLT can be used as follows. Process/product data is encrypted including time stamp and recipient address and transferred to a data marketplace based on the Distributed Ledger and can be purchased against payment (e.g. pay-per-use or subscription models). Confidence in the traded data is generated by the complete, traceable and unalterable storage of the data/transactions including time and location stamp. Furthermore, the consensus mechanism of the DLT contributes to the building of trust, since it requires agreement on a transaction in the network and thus enforces that actions within a DLT-based network cannot be performed by an entity, since transactions and actions in the network are reconciled and stored transparently [23]. Because the platform operator does not manage and host the selected Distributed Ledger on his own, the platform operator himself cannot access the data of the participants unnoticed (see Figure 11). If he were to do so, the transparency of the data marketplace for the data alliance would make the activity noticeable. The platform operator would thus significantly reduce the attractiveness of his data marketplace and therefore act against his own interests.
When selecting the underlying DLT, its individual characteristics must be taken into account. The characteristics of the application case should determine the choice of technology. For example, if the application case requires high scalability and transaction speed, DLTs from the category of Directed Acyclic Graphs (DAG) are more suitable than the classic Distributed Ledger Technology block chain.
4.4 From Data Monetization to Data Economy
Digital ecosystems are distributed, adaptable, and open socio-technical systems, in which properties of self-organisation, scalability and sustainability are oriented towards natural ecosystems. In this context, effects of competition and cooperation of different actors within the ecosystem play a central role [24].
4.4.1 Actors
The individual relevant actions for a monetization of manufacturing data (see 4.1) do not have to be performed by the same entity. They can also be carried out by partners with special know-how through cross-company value creation. As soon as data can be purchased, refined, and resold as a resource, a network of entities is created, which automatically evaluate data and translate it into monetary added value for other companies, thus opening up new value creation streams for themselves and others.
A data producer collects the data acquired by IoT/IIoT devices within its own company and takes only minimal steps itself to preprocess the data. The raw data is then linked to the context provided by the data producer and, if necessary, embedded in syntax-based models. Smaller and uncritical excerpts (exemplary and possibly synthetic data sets) are then made freely available in the data marketplace to provide potential buyers with insights into the type of data set.
Data service providers view the available data and can either buy and link data driven by their own innovation ideas or solve concrete needs of already identified customers. This business model describes the sale of data-based services that are available to all other participants in the network.
Refined data sets and AI-based models can be purchased by users and integrated into their production. Basic functions of the model as well as benchmarks for general performance are stored in the data marketplace. For users it is essential that the models are integrated into services that can be easily integrated into the existing infrastructure of the company and that have a high reliability.
The business model of the platform operator is to provide the technical infrastructure. It is characterized by the exercise of the trustee function, which guarantees the security of transactions and data as well as data integrity. Driving force of the platform operator is the expansion of the network by new partners who can be assigned to one of the participant categories mentioned above. An independent ecosystem based on the platform is developing.
4.4.2 Rating system
Participants in the network must be incentivized to provide data and services and to ensure their quality. Otherwise, a data marketplace degenerates into a kind of data dump [25], where it is not possible to distinguish between good and bad data/data services. The introduction of a rating system for the participant categories makes it possible to sort providers according to defined quality criteria. Conversely, the results of the quality ranking determine the attractiveness of their offers. The objectivity of the ranking system is essential to create fair conditions within the platform. For example, models can be used to evaluate data quality. The FAIR data principles are considered as guiding principles for making data discoverable, accessible, interoperable and reusable [26]. They provide guidance for scientific data management and data administration. The relevance of the FAIR data principles extends to all stakeholders in the current digital ecosystem [27]. Important aspects for the assessment of data quality are above all traceable responsibilities for the data, the reliability with which a data source produces new data, transparent versioning of data and the correct semantic embedding of data in a larger context. In addition, models offered must always include the data sets that were used to create the model [28]. By adhering to these principles, not only can data be better exchanged, viewed and processed, but models can also be clearly assigned to specific data sets, thus significantly increasing transparency within the data marketplace.
4.4.3 New value creation
The monetization of manufacturing data provides companies with new options for designing investments and business models. Investing in data that is valuable to a company, the planned purchase of data to expand its own models, the direct provision of models in a pay-per-use or AI‑as‑a‑service approach have an impact on existing business models as well as on the development of new ones. In a fully developed ecosystem, mechanisms automatically emerge after some time that set the prices for certain data according to quality standards and the informative value of the data.
One approach to pricing at the beginning of an ecosystem is the free provision of in-house data sets by a data producer. Providers of data services can then create a data-driven model, e.g. for predicting quality characteristics, and thus solve a concrete problem within (possibly third party) companies. This model can then be released for usage in exchange for incentives.
In order to remunerate the data producer for the data used, the producer can be paid a percentage of the added value of the model. If the developed model brings operational advantages for the data producer, the data producer can purchase or use the model itself (possibly at discounted conditions).
Such a model would on the one hand allow data producing companies to generate added value from their own data without a high ex ante investment risk in AI know-how. On the other hand, a situation would arise in which data service providers would have a large number of data sets at their disposal to implement an innovative combination of different data sets and develop new data services on this basis. If it is not possible to develop product-ready models on the basis of the data offered, there are no costs for the use of the data. On the other hand, data producers benefit directly if their data has contributed to the modelling of a product-ripe model. The number of companies willing to provide their data and make it “freely” available is a critical factor for the emergence of data-based innovations within the ecosystem [29].
Summary: The four main obstacles to monetization of manufacturing data are i) a lack of corporate strategies for data acquisition, ii) a lack of availability of AI know-how, iii) a lack of decentralized platforms allowing automated and specific exchange of data, and iv) a lack of clarity about the value of data and loss of know-how without adequate monetary reward through data sharing. The proposed solutions provide initial approaches to meet these challenges. The core of the solutions is, on the one hand, the decentralized storage and administration of data, so that no central entity has sovereignty over the data or the network. On the other hand, participation models can negate the ex ante investment hurdles in the area of data analysis. This creates incentives for companies to participate in an ecosystem for data monetization and enables the use of data-driven knowledge spillover effects.
5. Scenarios for Manufacturing Technologies
The abstract concepts of data monetization can be reduced to concrete scenarios that highlight the added value of the solutions presented. In the following, three scenarios are presented which vividly describe three basic mechanisms of data monetization. The scenarios were developed and discussed in a joint dialogue with experts from business and science in the fields of manufacturing, IT (software and hardware), and AI.
5.1 Selling secondary data: Product benchmark
Company profiles: Company A is a manufacturer of safety critical components using machine presses. Company B is a supplier of auxiliary and operating materials to ensure the smooth manufacturing of millions of metal components.
Problem description: Although Company A manufactures millions of components using a stable process, no part is like the other and several hundred parts are rejected in recurring patterns.
Cause: Company A produces 24 hours a day, 365 days a year. Due to seasonal and daily fluctuations in environmental parameters, such as hall temperature, humidity and others, the properties of the lubricant and the process change to a small extent, so that in unfavourable scenarios tolerances are violated and rejections occur. Due to the several times higher tool hardness compared to the material hardness, tool wear can be neglected.
Solution approach: An established parameter for evaluating the effectiveness of lubricants is the resulting process force. At the same time, however, the process force and the hall environment information is not data that endangers business. Company A could therefore sell this data package consisting of environmental data and process forces of rejected parts to company B. Company B would now be able to use this data to approximately reproduce the manufacturing conditions for its domain and thus optimize its lubricant product. Company B is willing to buy the data package because it expects to increase sales to Company A and strengthen customer loyalty by improving its lubricant. Company A is interested in an improved lubricant product because it can use internal monetization added value (reduction of rejects).
Expected added value through monetization: The primary added value for Company A is to develop new income streams through the sale of machine data. The other monetization effects for Companies A and B are linked with uncertainties and an R&D risk and are therefore not considered.
“The assessment of the sale of data to third parties is currently difficult to grasp in monetary terms. It is expected that based on the data, services will be derived which will lead to operational cost savings of at least 15 %”. — CEO Metalworkers
5.2 Buying third party data: Individualization of the material
Company profiles: Company A is again the manufacturer of safety-critical components. This time company B is a supplier of metal workpiece material.
Problem description: Although Company A produces millions of components using a stable process, no part is like the other and several hundred components are rejected in recurring patterns. Due to a new improved lubricant, the influence of operating and auxiliary materials can be neglected.
Cause: Company A continues to produce 24 hours a day, 365 days a year. Measurements on workpiece material samples have shown that material thickness and strength vary greatly from batch to batch and even along a batch. Due to the highly complex steel manufacturing process, it is for natural reasons not possible to produce a better material quality. Because Company A does not know the material thickness and strength, it cannot adjust the machine parameters accordingly and runs with an empirical machine setting, which does not have the desired effects from batch to batch.
Solution: Due to the specific process, it is not possible to measure material thickness and strength in process. Company A is also not willing to do so, as it is not an expert in material characteristics, but a manufacturer of millions of metal components. Company B, on the other hand, is a steel manufacturer and supplier to Company A. Company B has the necessary infrastructure and know-how to measure and interpret the material thickness and strength and to sell this data set to Company A. Company B is prepared to make this effort because it sees a risk that Company A might otherwise switch to another company. Company A can now adjust selected machine parameters for each batch and for each excessive fluctuation, so that material fluctuations can be compensated in the process.
Expected added value through monetization: The primary added value for Company A is that by purchasing data from third parties, its own processes can be improved in such a way that there is a reduction in scrap and thus an increase in manufacturing profits. The added value of Company B was discussed in Use Case 1 and further added value for both parties is linked to uncertainties that are not to be considered.
“The purchase of third-party data enables a reduction of time wasted by approx. 15 % and offers the additional potential to substitute storage space with productive space. This generates additional revenues and enables sales increases of at least 23 %.” — Manufacturing manager metalworking
“By purchasing data, the establishment of expert knowledge within the company is no longer necessary and also increases the efficiency of those involved in the process. This optimization of human resources increases the performance by 30 %.” — CEO Metalworkers
5.3 Innovate: Data-based service support
Company profiles: Company C is the manufacturer of machine presses, which company A, among others, has produced safety critical components with. Company B is another manufacturer of other safety-critical components and has received the machine press from Company C today.
Problem description: Company C has already installed hundreds of machine presses worldwide. Due to the size and complexity of the machine presses, installation and commissioning takes several days. Also, not everything runs as planned. This leads to problems and delays. For Company C, this results in significantly higher costs for assembly personnel and Company B suffers losses from lost profits due to the downtime.
Reason: Usually employees on assembly are on their own and do not have full knowledge of the location. Therefore, they often have to make spontaneous decisions without a sufficient information basis and without being able to assess the consequences.
Approach: Digital assistance systems are data-based systems that provide supporting information by means of virtual, augmented or mixed reality. This supporting information reduces the risk of making wrong decisions and can therefore reduce the implementation time on site. In order to set up such a system, Company C needs to log and analyze data about the implementation process. Companies A and B are willing to do this, because firstly, they are paid a one-time fee, secondly, they only have to provide anonymized data, and thirdly, they themselves will benefit from it in the future, if they want to purchase a new machine.
Expected added value through data monetization: The primary added value for Company C is the reduction of assembly days required and a significant increase in customer service quality. Both are clear market advantages compared to competitors. Other added values will not be considered further in this paper.
“External technical service and maintenance are essential components of successful manufacturing. By analyzing third-party data and projecting it onto the company’s own organizational and infrastructure, a neutral and open system evaluation is possible. It is calculated that such approaches will lead to cost savings of at least 350,000 €/a.” — Head of Maintenance Metalworking
“The approach enables anticipatory control of internal processes and allows reliable planning of processes, which can significantly reduce unnecessary downtime of manufacturing machines. An initial estimate based on empirical values forecasts annual cost savings of 475,000 €.” — Plant Manager Metalworking
6. Conclusion
The monetization of manufacturing data can lead to data-driven knowledge spillover effects for participants in a data alliance through the use and exchange of (aggregated and transformed) manufacturing data. This allows existing business processes to be optimized and innovations to be created. The implementation of a targeted data monetization strategy promises to open up new income streams and gain new market shares through digital, innovative and progressive business models. A data monetization strategy may look as follows.
At the beginning of the monetization of manufacturing data, data-producing companies should look at their own database, ensure the context of the data, and cleanse the data collected. Once all incomplete or inaccurate data points from a data set have been identified and corrected, the data is in an optimal state for internal monetization. Vertical process analysis within a company makes it possible to gain new insights into manufacturing or business processes, reduce costs, optimize products and minimize risks with little effort.
Building on the success of internal data monetization and the resulting lower inhibition threshold, the next step can follow: Data exchange with another data producer and the formation of so-called data alliances. Overcoming the threshold of usability requires the use of AI models. To do so, the following four steps must be worked out within the data alliances:
- Design of a data collection, decentralized and edge-based infrastructure
- Clarification of issues and requirements in the area of data privacy, data integrity and data sovereignty
- Definition of the monetization strategy
- Development of pricing models and participation options
Finally, there are five steps for manufacturing companies to define and implement a data monetization strategy:
- Identify appropriate data sources and integrate them into a site-/company-wide interoperable data system
- Develop a strategy for the release of company data to third parties, including, for example, the scope of the data and, above all, the context in which the data is located
- Identify suitable partners for the formation of data alliances, shaping and continuously expanding these partnerships
- Define innovative participation models for the usefulness of their data
- Integrate the partnerships developed into a platform-based ecosystem in the form of a data marketplace that enables data-based innovation through openness to data exchange between companies from different industries, thereby economically strengthening data producers, platform operators and data service providers
For companies without internal AI know-how, the use of external resources is essential for identifying the economic potential of their own data sets, because the risk of bad investments is not borne by the company but outsourced to data service providers. These providers can develop innovative data products within a broad ecosystem. Through sale or alternative participation models, purely data-producing companies can also participate in the economic success of the resulting data products.
References
[1] Carlino, G. A. (2001). Knowledge spillovers: cities’ role in the new economy. Business Review Q, 4(1), 17–24.
[2] Glaeser, E. L., Kallal, H. D., Scheinkman, J. A., & Shleifer, A. (1992). Growth in cities. Journal of political economy, 100(6), 1126–1152.
[3] Huber, F. (2012). Do clusters really matter for innovation practices in Information Technology? Questioning the significance of technological knowledge spillovers. Journal of economic geography, 12(1), 107–126.
[4] Castillo, V., Figal-Garone, L., Maffioli, A., Rojo, S., & Stucchi, R. (2016). The effects of knowledge spillovers through labor mobility.
[5] Pennekamp, J., Glebke, R., Henze, M., Meisen, T., Quix, C., Hai, R., … & Epple, A. (2019, May). Towards an Infrastructure Enabling the Internet of Production. In 2019 IEEE International Conference on Industrial Cyber Physical Systems (ICPS) (pp. 31–37). IEEE.
[6] Li, G., Wang, J., Wu, J., & Song, J. (2018). Data processing delay optimization in mobile edge computing. Wireless Communications and Mobile Computing, 2018.
[7] Fedkenhauer, T., Fritzsche-Sterr, Y., Nagel, L., Pauer, A., & Resetko, A. (2017). Datenaustausch als wesentlicher Bestandteil der Digitalisierung. PricewaterhouseCoopers. URL: http://www.industrialdataspace.org/ressource-hub/publikationen. Access: 31.03.2020.
[8] World Economic Forum. (2020). Share to Gain: Unlocking Data Value in Manufacturing. White paper.
[9] Yan, J., Meng, Y., Lu, L., & Li, L. (2017). Industrial big data in an industry 4.0 environment: Challenges, schemes, and applications for predictive maintenance. IEEE Access, 5, 23484–23491.
[10] Pennekamp, J., Henze, M., Schmidt, S., Niemietz, P., Fey, M., Trauth, D., … & Wehrle, K. (2019, November). Dataflow Challenges in an Internet of Production: A Security & Privacy Perspective. In Proceedings of the ACM Workshop on Cyber-Physical Systems Security & Privacy (pp. 27–38).
[11] Zhong, R. Y., Xu, C., Chen, C., & Huang, G. Q. (2017). Big data analytics for physical internet-based intelligent manufacturing shop floors. International journal of production research, 55(9), 2610–2621.
[12] Chen, S., Shi, R., Ren, Z., Yan, J., Shi, Y., & Zhang, J. (2017, November). A blockchain-based supply chain quality management framework. In 2017 IEEE 14th International Conference on e-Business Engineering (ICEBE) (pp. 172–176). IEEE.
[13] Wu, S. (2013). A review on coarse warranty data and analysis. Reliability Engineering & System Safety, 114, 1–11.
[14] Cui, Z., Chen, W., & Chen, Y. (2016). Multi-scale convolutional neural networks for time series classification. arXiv preprint arXiv:1603.06995.
[15] Liu, C., Li, Y., Zhou, G., & Shen, W. (2018). A sensor fusion and support vector machine based approach for recognition of complex machining conditions. Journal of Intelligent Manufacturing, 29(8), 1739–1752.
[16] Kłosowski, G., Gola, A., & Świć, A. (2016). Application of fuzzy logic in assigning workers to production tasks. In Distributed Computing and Artificial Intelligence, 13th International Conference (pp. 505–513). Springer, Cham.
[17] Mahendrawathi, E. R., Arsad, N., Astuti, H. M., Kusumawardani, R. P., & Utami, R. A. (2018). Analysis of production planning in a global manufacturing company with process mining. Journal of Enterprise Information Management.
[18] Jüttner, U., Windler, K., Schäfer, A., & Zimmermann, A. (2017). Design von Smart Services–Eine explorative Studie im Business-to-Business-Sektor. In Dienstleistungen 4.0 (pp. 335–361). Springer Gabler, Wiesbaden.
[19] Meisel L., Spiekermann M. (2019). Datenmarktplätze — Plattformen für Datenaustausch und Datenmonetarisierung in der Data Economy. Fraunhofer-Institut für Software- und Systemtechnik ISST.
[20] Schürmann, H. (2017). Datenanalysten sind rar. https://www.vdi-nachrichten.com/karriere/datenanalysten-sind-rar/. Access: 31.03.2020.
[21] Peters, R. (2010). Internet-Ökonomie. Springer-Verlag.
[22] KPMG, Bitkom. (2015). Mit Daten Werte schaffen. Report 2015. https://www.bitkom.org/sites/default/files/file/import/KPMG-Bitkom-Research-Studie-MDWS-final-2.pdf. Access: 31.03.2020.
[23] Morkunas, V. J., Paschen, J., & Boon, E. (2019). How blockchain technologies impact your business model. Business Horizons, 62(3), 295–306.
[24] Briscoe, G., & De Wilde, P. (2006, December). Digital ecosystems: evolving service-orientated architectures. In Proceedings of the 1st international conference on Bio inspired models of network, information and computing systems (pp. 17-es).
[25] Freiknecht, J. (2018). Big Data in der Praxis: Beispiellösungen mit Hadoop und NoSQL. Daten speichern, aufbereiten, visualisieren. Carl Hanser Verlag GmbH Co KG.
[26] Freiknecht, J. (2014). Big Data in der Praxis: Beispiellösungen mit Hadoop und NoSQL. Daten speichern, aufbereiten, visualisieren. Carl Hanser Verlag GmbH Co KG.
[27] Mons, B., Neylon, C., Velterop, J., Dumontier, M., da Silva Santos, L. O. B., & Wilkinson, M. D. (2017). Cloudy, increasingly FAIR; revisiting the FAIR Data guiding principles for the European Open Science Cloud. Information Services & Use, 37(1), 49–56.
[28] Gleim, L., Pennekamp, J., Liebenberg, M., Buchsbaum, M., Niemietz, P., Knape, S., … & Brecher, C. (2020). FactDAG: Formalizing Data Interoperability in an Internet of Production. IEEE Internet of Things Journal.
[29] Zillner, S., Becker, T., Munné, R., Hussain, K., Rusitschka, S., Lippell, H., … & Ojo, A. (2016). Big data-driven innovation in industrial sectors. In New Horizons for a Data-Driven Economy (pp. 169–178). Springer, Cham.
You have questions? Please contact us!
senseering GmbH
The senseering GmbH is a company founded in September 2018 that was awarded the RWTH Aachen University Spin-Off-Award. The core competence of senseering GmbH is the development and implementation of systems for the digitalization and networking of industrial and production facilities. Likewise, senseering GmbH advises on strategic corporate issues, in particular digital transformation, distributed-leger technologies, edge vs. cloud computing architectures for AI-based real-time control of industrial processes, digital business model innovation and the introduction of digital business processes such as home office, Azure or Microsoft365. Senseering is one of the winners of the first and largest AI innovation competition of the BMWi with the project www.spaicer.de.
Daniel Trauth (CEO) | www.senseering.de | E-Mail: mail@senseering.de
