Due to the increasing competitive pressure of globalization, the production of high-quality products is crucial for the long-term success of a company. In order to guarantee the delivery and transfer of faultless products, it is therefore essential to ensure quality. In the classical sense, this is in turn associated with high testing volumes, so that on the one hand high investments are made for testing equipment and on the other hand high testing costs are incurred. Especially against the background of reducing throughput times, increasing flexibility and increasing efficiency, strategies for reducing the inspection scope are economically attractive compared to the acquisition of further systems and test equipment. With a strong brand awareness and in order to avoid high recall actions, this was however in the past a rather unusual measure for traditional companies. Due to the increasing spread of modern information and communication technologies, however, the use of data-driven methods, e.g. the use of data mining methods, in industrial production environments is also favoured in this context and, in addition, high economic potentials can be leveraged.
The use of data mining in production offers great potential for the development and integration of strategies for the optimization of products and production processes. By applying statistical methods to structured and unstructured data, previously unknown patterns and laws can be extracted and new knowledge can be generated. This enables the creation of forecast models for data-based and computer-aided prediction of future events.
This presentation will show how the use of data mining in electronics production can relieve X-ray inspection by predicting quality. On the basis of collected process data, prognosis models are trained, which allow a prediction of the expected X-ray result of the printed circuit boards. Early knowledge of the product quality to be expected enables early control interventions, so that on the one hand the inspection scope is reduced and on the other hand additional added value of defective products is prevented.
Increased price and competitive pressure as well as initiatives to increase energy and resource efficiency present the beverage industry with major challenges for rationalization. In the presentation, the possibilities and limits of existing approaches to data-based process optimization will be presented and it will be shown that new approaches to data analysis are required for biochemical processes with complex combinations of different influencing variables. For this reason, the DaPro research consortium was established to explore data driven process optimization in the beverage industry based on machine learning.
Accurate freight flow predictions allow better shift planning and resource allocation. This presentation describes how RapidMiner can be used to integrate predictive analytics into an existing IT architecture built on diverse platforms and including third-party providers. The generated forecasts allow for precise freight flow predictions based on location, day of the week, and even down to a shift level, enabling Lufthansa Industry Solutions’ customers to precisely schedule shifts and their required manpower. This presentation gives an overview of the software architecture, the integration of RapidMiner into the productive environment with Hadoop, the concept of the predictive model built, as well as an excursion into test automation with RapidMiner.
Martina Ringeln
Deputy Director of Quality Management
The analysis of field data as a basis for continuous optimization of products and processes is one core area of a sustainable quality management. In the field of consumer products especially the evaluation of customer and service data to identify defects is complicated by an unstructured and heterogeneous presence of data.
This presentation gives an insight how text mining and data mining are applied to field data at Miele and how Miele is going to improve the staff deployment for handling unstructured texts. The goal of Miele is to reduce the amount of screen work with a consistent high quality of analysis in the context of Industry 4.0 and Big Data in order to save time for other work activities.
In order to leverage the power of data science, it has been necessary to ensure the technology stack meets the needs of our analyst community and that the correct technical skills are in place to compliment the domain-knowledge across the functions. This presentation explains some of the challenges faced on the journey to becoming a more analytical organisation and the success stories that data science has brought.
Prof. Dr.-Ing, Jochen Deuse
Head of the Institute of Production Systems (IPS)
The modern industrial production environment receives strong impulses through an ever increasing use of data science methods for optimization purposes. In particular, the consideration of three essential success factors is of great importance for the efficient implementation of such industrial data science projects. Firstly, the project-internal procedure must be aligned with a structured procedure model, such as the CRISP-DM. Secondly, the project team has to be sufficiently interdisciplinary, for example ensuring sufficient domain knowledge. Thirdly, it is necessary to ensure sufficient data quality and quantity for the purposes of analysis, with the support, for example, of using so-called maturity models.
Prof. Dr. Katharina Morik
Head of Artificial Intelligence Unit
AI has raised some interest, recently. The driving force behind this hype is machine learning, in particular deep learning. This talk sketches the current hot topics and then focuses on what is relevant for applications in industrial processes.
Ralf Klinkenberg
Co. Founder, Head of Data Science Research
This presentation provides an overview of industry applications of machine learning and predictive analytics in the automotive, aviation, chemical, manufacturing, pharmaceutical, steel, and other industries covering the following use cases: Predictive Maintenance: Predicting and Preventing Machine Failures before they happen; Prediction, Prevention, and Resolution of Critical Situations in Continuous Production Processes; Product Quality Prediction in early stages of the production process; Optimization of Production Processes; Optimization of Mixture of Materials or Ingredients; Assembly Time and Assembly Plan Prediction for New Product Designs; Demand Forecasting and Price Forecasting; Web Mining and Information Extraction for Semi-Automation of Price Quote Generation and Purchase Processes; Web Mining and Information Extraction for Market Intelligence, Trend Monitoring, and Competitive Intelligence; Semi-Automated Data Augmentation from Internal and External Data Sources. These use cases cover both, industry deployments as well as new application use cases from the RapidMiner Research Lab.
Speakers from 2017
Speakers and presentation abstracts from the last Industrial Data Science Conference (IDS 2017). In 2017 speakers from ABB, Achenbach Buschhütten, Arcelor Mittal, BMW, Daimler, Deutsche Edelstahlwerke, Lufthansa Industry Solutions, Miele, VW, and others described machine learning application use cases in their given industry. In order to see the presentations, i.e. presentation titles, descriptions, slides, videos, and speaker biographies, please click on Talks in the top menu and then on the image next to the title of the presentation you are interested in.
Marcel Dix, ABB
The application of big data and data analytics for optimizing production and mainte- nance processes is one of the most important use cases for industrial IoT. The talk introduces ABB’s development process for data driven services. The development process is essentially organized in four steps: 1. Development of value propositions and deduction of relevant analytical questions 2. Collection, investigation, and preparation of available data 3. Data analysis and model training 4. Roll-out and deployment of Big Data and analytics based services. The process is illustrated based on an example from chemical industries, but many aspects can be transferred to other applications.
Dr. Olga Erohin, Miele
The application use cases of data mining in industrial environment are manifold. They vary from condition monitoring and quality prediction to sequence optimiza- tion and text mining for warranty analysis. In this presentation, we describe the knowledge discovery process specialized in time data management as application field and two use cases of assembly time prediction for manual assembly processes.
Roger Feist, Achenbach Buschhütten
When algorithms and solutions have been verified in laboratory environments, it’s time to pave the way for their real field application. Here Optilink® becomes a product with interesting perspectives. Starting with a bottom-up approach, Optilink® is a tool chain to acquire, transmit, and analyze huge amounts of real-time- and historical-data of machinery in a safe and scalable cloud environment. Automated feature extraction is possible by the use of time- and data-driven events. This talk will present the tool chain, underlying concepts, and some examples for applications.
Dr. Gabriel Fricout, Arcelor Mittal
Data mining processes have proven to be very valuable for addressing industrial issues such as understanding defect crisis. In classical data mining procedures, only ‘single value’ variables are considered, meaning that one individual, in the steel industry typi- cally one coil, is characterized by average values of many process parameters (composition, temperature, speed, strengths, tractions, composition, etc.), which will be used to predict, forecast, or estimate unknown properties about the product, such as the probability of defect occurrence for instance. However, in many situations, the available information is much wider: Many sensors continuously register information about the product and the process. This talk focuses on presenting both methodology and tools to perform data mining using time series from the steel manufacturing pro- cess sensors. The methodology involves several steps of data preparation, statistical modelling and training (based on shapelet and deep learning methods), performance evaluations, and knowledge capitalization. Results are illustrated on real data from the steel production process trying in particular to forecast as early as possible the risk of product non-quality. Tools involved and developed are technological choices based on various available pieces of software (MongoDB, RapidMiner, Kasem). All the develop- ments are conducted within the PRESED RFCS framework.
Ferdinand Klenner, BMW
In the context of Industry 4.0, flexible manufacturing systems are the conceptual basis for highly flexible and efficient high-volume manufacturing. But the inherent complexity of these systems and the lack of transparency impede the further increase of their efficiency. If the number of machine breakdowns exceeds the resources of operators and maintenance, it is necessary to guide the employees to the most critical machines. Based on the approach of the Theory of Constraints, the most critical machine is called bottleneck and restricts the throughput of the entire system to any given time. Thus, an effective decision support system for production employees needs a real-time identification of momentary bottlenecks. Furthermore, a bottleneck prediction can support the employees to plan actions in advance and prevent uncoordinated firefighting. Hereby, production data, modern IT-infra- structure, and data mining applications enable the development of real-time identification and prediction of bottlenecks in such complex, flexible manufacturing systems.
Dr. Fabian Werner, Lufthansa Industry Solutions
How do data scientists and engineers best bring their expertise into a productive state? – Lufthansa Industry solutions uses a standard approach to operationalize data analytics at organizations. We will give a general overview over analytics use cases at Lufthansa and show some practical use cases from the airline industry such as the prediction of arrival times of aircrafts or whether passengers show up in or- der to take their flight. This includes some words on techniques and technology which are being used in order to tackle these exciting cases. In our airline, a proto- type is in evaluation right now - when an aircraft worldwide goes airborne, our pro- cess pulls the data for that flight and produces a prediction. Furthermore we devel- oped a Web App for measuring the performance, an evaluation process, and a su- pervisor which regularly checks whether or not the model building, rolling window, and prediction processes work as expected. By using our predictions, employees of the Hub Control Center may detect short-timed connections in advance and cause counter measures like a direct transfer of passengers from their inbound flight to their outbound flight. We plan to extend the model in a way that it will incorporate live geolocation data about flights all over the world.
Ralf Kretschmer, Miele
The application use cases of data mining in industrial environment are manifold. They vary from condition monitoring and quality prediction to sequence optimiza- tion and text mining for warranty analysis. In this presentation, we describe the knowledge discovery process specialized in time data management as application field and two use cases of assembly time prediction for manual assembly processes.
Dr. Regina Wallis, Daimler
The application of digital assembly planning tools leads to a large amount of data documenting product development and assembly planning results along the product emergence process. The reuse and adaptation of existing planning data can accelerate future planning processes. The presented approach uses data mining techniques to generate a suitable assembly work plan from existing planning data as a starting point for a new assembly planning task. Classification and clustering techniques are applied to both product and process data in order to identify their correlations.
Prof. Dr. Katharina Morik, TU Dortmund University
She describes challenges and approaches for the analysis of very large data sets and high-dimensional data under resource constraints, for example for large and fast data streams and for distributed data. Large fast data streams are for example sensor measurements in the industry for predicting issues in the production process in advance and traffic sensors data streams for predictive traffic planning. Solutions leverage a broad range of methods including statistical learning algorithms like Support Vector Machines (SVM) to graphical models as well as modular and flexible integrated advanced predictive analytics platforms.
Ralf Klinkenberg, RapidMiner
After a short description of the state, challenges, barriers, use cases, and opportunities of Industrial Data Science and of the Cross- Industry Standard Process for Data Mining (CRISP-DM), which is used as a redline through this event, we provide a short overview over the data science use cases presented at IDS 2017, whose presentation order reflects the steps in CRISP-DM. IDS 2017 focuses on industrial data science applications and best practices, i.e. hands-on experience how to best use data to solve problems in industry and to leverage opportunities. Learn from and discuss with industry experts.
Prof. Dr.-Ing. Jochen Deuse, Institute of Production Systems
The challenges and requirements an industrial engineer has to face in his daily busi- ness have significantly changed in the context of Industry 4.0. While the core tasks of the Industrial Engineer are rooted domain specifically in work and motion studies and work system design, his methodological spectrum is increasingly changing from using basic descriptive statistics towards digital manufacturing, factory physics, and machine learning.
