HILTON MUNICH PARK
12-13 JUNE, 2018

WHERE AI MEETS IoT

Agenda


Predictive Analytics World for Industry 4.0 - München - Day 1 - Tuesday, 12th June 2018

8:00 am
Registration and Breakfastsnack
9:15 am
Keynote

Fuelled by the success of Deep Learning in image, video and speech recognition, neural networks (NN) and other Artificial Intelligence algorithms are being widely tested in industry practice. In our talk, we will show company case studies where AI has shown super-human performance in descriptive, diagnostic and predictive analytics. Using image recognition, we trained AI to recognise complex time series behaviour for fully automatic forecasting model selection mimicking experts in econometrics and descriptive analytics. In diagnostic analytics, we applied AI to outlier detection for the predictive maintenance of machines, in predictive analytics AI predicts future sales for production planning under a multitude of external influences including price and promotional information as well as bank holidays and weather effects, and in prescriptive analytics by learning cost-efficient inventory levels directly from the data.

However, while companies are busy leveraging selected AI in their industry practices, research is already moving further ahead to put AI to even wider use in replacing human experts. We will finish the talk by introducing the next generation paradigms such as transfer learning, meta learning, and active learning, which will enable AI to selectively learning from data, or to acquire more data to derive better decisions autonomously, in order to learn complex multi-purpose tasks beyond a single application domain. 

Session description
Speaker
Prof. Dr. Sven CroneLancaster Research Centre for Forecasting
Director
Lancaster Research Centre for Forecasting
10:10 am
Coffee Break
10:40 am
Case Studies: Logistics & Mobility

Embracing Innovations Continental Tires always is looking for new business models to complement it's almost 150 years of experience as a tire manufacturer. With 360° many services around the tire world entered the stage. Currently, IoT methods are helping to get a deeper understanding of what is happening to the tire in real life situation in the field in every second. In this case study Dubravko and Christian will show how they collect data from the field and apply advanced analytics to the collected data.

Session description
Speakers
Dubravko DolicContinental Reifen
Continental Reifen Deutschland
Christian KarraschContinental Reifen
Continental Reifen Deutschland

We demonstrate a case study for responsible AI on the example of Swiss Postauto's autonomous Smart Shuttle.
Machine learning / artificial intelligence (AI) systems need special approaches to be de-biased, interpreted, and trusted. PwC develops services based on LIME, QII and other assessment techniques for black-box systems to build trust in AI. We analyze the influence of individual input factors and search for bias, possibly leading to overreaction on specific inputs. We determine whether algorithms are fully trained and identify potential risks of developing an algorithm into production. Our case study demonstrates how to set up trustful AI systems in organizations.

Session description
Speaker
Dr. Christian SpindlerPwC
Manager / Data Scientist
PwC
Deep Dive:

Reciprocating compressors are one of the most serious and expensive asset in the plant, these types of machines can provide higher compression ratio than similar axial or centrifugal compressors, however reciprocating compressors are more costly to maintain and suffering from high essential drawback. In spite of the criticality and importance of the reciprocating compressor, they are sometimes unobserved by condition monitoring team, simply because their typical and primary PDM tool ( FFT portable vibration analyser ) which is routinely used on rotating equipment is not well suited for reciprocating machines, and have been unsuccessfully monitoring the reciprocating machines for many years, therefore the overall machine health is frequently ignored and not diagnosed correctly until damage occurs and it became too late to save the machine from failure. At the same time experience and statistical studies show that reciprocating compressors almost consume times as maintenance cost of counterpart centrifugal or axial machines in the plant, consequently many companies spend more money on maintenance, repair and down times than it would cost to predict failure and eliminate the root cause on these critical machines, that approach could allow for maintenance to plan their activities around CBM rather than follow run to failure strategy.


The main goal of Reliability and condition monitoring team is not only to protect their reciprocating machines from catastrophic failure by using tradition protection system and vibration overall trending, but also predict, diagnose any abnormality in the equipment and sustain reliability which could be achieved by monitoring all machine components simultaneously, recording the performance, operation parameters, Suitable vibration software, overall health and lubrication as well.

Reciprocating compressor condition monitoring systems must achieve the level of condition monitoring that centrifugal equipment users have implemented for decades. Amir encourages top management to start investing in close monitoring their existing reciprocating compressors by implementing the new technologies which are available now after calculating the benefits on machine uptime and reliability.

Start ask your condition monitoring and reliability team, if they are confident enough with their current existing systems for monitoring reciprocating compressors as they are the same toward centrifugal ones, and you might be shocked that your receips is not detectable and monitored well, it’s time to take advantage of the technology that is becoming available.

Session description
Speaker
Amir BasyouniEprom
Head of Condition Monitoring
EPROM
11:40 am
Short Break
11:45 am
Case Studies: Logistics & Mobility

Data is at the heart of everything Uber does. Terabytes of data is streamed, processed and consumed everyday to make important decisions. Thus, it is crucial that the data infrastructure is fast, efficient and reliable. But sometimes, due to various reasons, quality of the data is compromised and as a result, decision makers see wrong information. Now at Uber's scale, each and every wrong decision can have disastrous impact on rider experience or earnings of drivers. So, how can we make sure that the data which is being streamed into databases is of utmost quality? We are using machine learning techniques to gather stats on data quality and creating an anomaly detection system to inform decision makers in case of degraded data quality and alerts engineers about possible causes for faster mitigation.

Session description
Speaker

Cargonexx (www.cargonexx.de), an European digital trucking network with over 40k trucks for long distance trucking, uses AI techniques to run their network. Automated pricing is delicate when it comes to dynamic changing market prices, especially when current prices are mainly based on countless human negotiations. In this presentation Alwin and Can are presenting the challenges and brief solutions approaches on their way to develop the pricing algorithm for Cargonexx, which is used to derive the current market price for each individual transport request in real-time. The proposed price must reflect the current market situation on both market sides as precisely as possible, to minimize the economic risk for Cargonexx. Dispatching processes are also optimized by AI. Algorithms determine the best load-carrier combination and also plan vehicle tours ahead of the actual booked loads, based on predicted demand. The focus  of this presentation will be on Cargonexx’s operational challenges and the predictive techniques applied to solve the specific needs.

Session description
Speakers
Can AkinCargonexx GmbH
COO
Cargonexx GmbH
Dr. Alwin HaenselHaensel AMS
Business Analytics & Founder
HAMS
Deep Dive:

The idea behind industry 4.0 is complete transparency along the value chain. Main requirement for achieving this, is a holistic data management process in and between companies of interlinked value chains. Focus hereby should be the proactive data management between supply chain partners. The presentation will address this topic beginning with its potentials for the supply chain and following with findings of a comprehensive empirical study about the current status of cross-company data management. Finally, based on the study a multilevel phase model for cross-company data management will be presented to derive concrete recommendations for further actions.

Session description
Speakers
Christoph HeinHendricks, Rost & CIE
Business Intelligence Consultant
Hendricks, Rost & CIE
Prof. Dr. Wanja WellbrockHeilbronn University, Faculty of Management and Sales
Professor for Procurement
Heilbronn University, Faculty of Management and Sales
12:45 pm
Lunch Break
2:00 pm
Case Studies: Logistics & Mobility

Predictive Maintenance allows reducing plant shutdowns by detecting incipient failure and replacing unplanned by planned maintenance. Failure prediction models are built using historical process data labeled using fault events. However, when setting up a model usually there are not sufficient fault events available to reliably train the model. In this session you will see an automotive production use case, where the model is retrained automatically whenever a new fault has occurred or a false alarm has been raised. Maintenance log data is classified automatically to identify suitable labels for process data and determine times for retraining so that retraining can be carried out automatically without operator interaction.

Session description
Speaker
Dr. Christoph PaulitschSiemens
Senior Key Expert Data Analytics and Condition Monitoring
Siemens

Data analysis and machine learning are the key technologies when it comes to optimizing processes and products in the digital environment. Especially in the automotive industry, the digitalization has transformed the vehicle from a purely mechanical product into a highly complex, software-technical system. Increased digital complexity poses new challenges to quality assurance in automobile production. Based on vehicle and diagnostic data from vehicle manufacturing, methods for the early detection of anomalies and weak points can be developed to make production processes more stable and efficient. The presentation will show the implementation of such a data mining project in automotive manufacturing at Audi.

Session description
Speaker
Dr. Roland StoffelDSA Daten und Systemtechnik GmbH – SKYLYZE
Data Scientist
DSA Daten und Systemtechnik GmbH – SKYLYZE
Deep Dive:

In this session, "Augmenting Business Intelligence at the Edge for Industry 4.0," I will talk about how AI and machine learning are unlocking unprecedented business value from sensor data. The cycle of exploratory visual analytics, numerical encoding and embedded rules/models enables real-time surveillance and statistical process control. In combination with digital twins, this cycle supports understanding and fine tuning on software representations of physical assets. Combining sensor data with visual analytics tools lets operators see patterns of equipment productivity, degradation and stoppages. The embedding of numerical models, encoding such patterns- enables continuous surveillance, decision support and automated interventions when degradation is detected. Such technologies are augmenting business intelligence at the edge, driving operational efficiency and ROI.

Session description
Speaker
Varun KhandelwalTIBCO SOFTWARE
Solutions Architect
TIBCO SOFTWARE
3:00 pm
Coffee Break
3:30 pm
Case Studies: Logistics & Mobility

Our customer produces diagnosis-devices to support mechanics during repair of cars with analysis of fault codes and parameters. However, so far mechanics must rely on manual interpretation of the parameters and thus successful repair is often time-consuming and requires a lot of expertise.
To support mechanics, we developed analytical models that extract dependencies between frequently measured parameter combinations. During diagnosis new measurements are compared to the trained model and parameters with anomalies from expected behavior are automatically identified. Application of these models will support the mechanics during interpretation of car-diagnosis results and thus will lead to efficient diagnosis and repair.

Session description
Speaker
Rico Knapperanacision
Chief Data Scientist
anacision

Most of today's machine learning tasks deal with predicting atomic values. In Classification, for each object a class is predicted, i.e. a category assigned. In Regression, for each object a numerical value is predicted. Both are simple predictions in the sense that they only predict a single value per object. Most machine learning algorithms are also simple in the sense that they can only handle simple linear feature vectors. We propose a complex problem: Given complex hierarchical 3D designs of new products, predict assembly times and automatically generate assembly plans, i.e. sequences pf assembly steps to manufacture these products.

The solution was validated by predicting assembly plans for new truck engine components at Daimler Trucks. In this Case Study Ralf describes how to use machine learning to automatically predict assembly times and assembly plans for new complex product designs. This enables car makers and other manufacturers to accelerate the product design and assembly planning process, increasing the agility of the company and reducing the initial costs for new products.

Session description
Speaker
Ralf KlinkenbergRapidMiner
Founder & Head of Data Science Research
RapidMiner
Deep Dive:

Edge computing and the Internet of Things bring great promise, but often just getting data from the edge requires moving mountains. Let's learn how to make edge data ingestion and analytics easier using StreamSets Data Collector edge, an ultralight, platform independent and small-footprint Open Source solution written in Go for streaming data from resource-constrained sensors and personal devices (like medical equipment or smartphones) to Apache Kafka, Amazon Kinesis and many others. This talk includes an overview of the SDC Edge main features, supported protocols and available processors for data transformation, insights on how it solves some challenges of traditional approaches to data ingestion, pipeline design basics, a walk-through some practical applications and the integration with other technologies such as Streamsets Data Collector, Apache Kafka, and Hadoop.


Session description
Speaker
Dr. Guglielmo IozziaOptum UnitedHealth Group
Big Data Delivery Lead
Optum UnitedHealth Group
4:30 pm
Short Break
4:35 pm
Special
The Session Description will be available shortly.
Session description
Speaker
Simon Schneider42ai
Co-Founder
42ai
Deep Dive

Fraunhofer IISB is working towards zero-defect manufacturing with key players in the semiconductor industry. As in other industries, manual interventions by equipment operators have influences on product quality. Felix presents guidance for gathering and analyzing data with regard to manual interventions in mass manufacturing from a joint project with Infineon. Based on the wire bonding process he demonstrates how manual interventions by equipment operators are optimized to increase process stability. Whereas nowadays setup failures can only be derived by cost-intensive quality tests during equipment downtime, classification algorithm enables failure prediction by existing machine data: A role model to support equipment operators by exploiting available data!

Session description
Speaker
Felix KlingertFraunhofer IISB
Data Scientist
Fraunhofer IISB
5:30 pm
Reception and Networking at Exhibition Hall
7:00 pm
End of the first conference day

Predictive Analytics World for Industry 4.0 - München - Day 2 - Wednesday, 13th June 2018

8:00 am
Registration and Breakfastsnack
9:00 am
Keynote:

Working with data is different and challenges companies in a variety of ways. Installing a data lab alone is not enough, and it takes more than just data scientists and deep learning to generate sustainable business value from data & AI. In his keynote, Klaas will talk about the key ingredients for building up and developing enterprise-ready data & AI teams and spaces, but also about the toxic substances to destroy them. He will re-think approaches, share experiences and next practices. It is not about skills, technologies, labs or CDOs. It is about people, spaces, structures and (just) software in highly dynamic – it's about "fractal" – systems. Klaas will present how he built several data labs and is currently in the process of further developing and "mainstreaming" some of them in the direction of a factory, a hub or platform.

Session description
Speaker
Klaas Willhelm BollhöferBirds on Mars
Founder & AI Thinker
Birds on Mars
9:45 am
Coffee Break
10:15 am
Case Studies: Manufacturing, Energy & Health

Imagine your first day in a new industrial data science job. Rather than stats, code and models, you soon realise that trawling the corporate intranet for contacts and identifying promising opportunities for analytics becomes your primary activity. Add to that a mix of disconnected, organically grown data hoarding activities, along with half-finished projects sold by the usual snake-oil vendors. That's Boris in late 2017. In this presentation, he will aim to summarise his key learnings from the first 6 months at Merck's Chief Digital Office.

Session description
Speaker
Dr. Boris AdryanMerck
Expert for IoT & Data Analytics at the Chief Digital Office
Merck KGaA

Siemens Healthineers is the leading provider of medical imaging systems which are sending IoT data to a central analytics platform. This is used for predictive maintenance activities in the Customer Service division. In the talk we will give an overview of the use case and the analytical platform as well as some details on modelling methods used. These range from models driven by expert input to generic sequence mining models to deep recurrent networks to model signals over time.

Session description
Speaker
Dr. Tobias HippSiemens Healthineers
Data Scientist
Siemens Healthineers
​Deep Dive:

Neural networks have been proven to be universal approximators but this still leaves the identification task a hard one. Beside using data we should focus our attention on the underlying structure of our subject of interest. In case of dynamical systems this is time, leading us to state space models and recurrent neural networks. After an introduction of small (open) dynamical systems you will study dynamical systems on manifolds. Here manifold and dynamics have to be identified in parallel. You will move on to large (closed) dynamical systems with hundreds of state variables and will combine causal and retro-causal models of the observations. This combination leads to an implicit description of dynamical systems on manifolds. Finally you will discuss the quantification of uncertainty in forecasting. In our framework the uncertainty appears as a consequence of principally unidentifiable hidden variables in the description of large systems. Together with the mathematical concepts you will see applications in economics and engineering.

Session description
Speaker
Dr. Hans-Georg ZimmermannFraunhofer Institut
Senior Research Scientist
Fraunhofer Gesellschaft
11:15 am
Short Break
11:20 am
Case Studies: Manufacturing, Energy & Health

Rotorcraft vibration contains comprehensive information from multiple sources about the health state of a mechanical system. Decoding this information is key for the assessment of the machine health status and the identification of anomalies. Airbus Helicopters uses classical and advances techniques for the detection and diagnosis of incipient component faults based on vibration data collected during helicopter operations. Vibration data is capable of providing insights into the general assembly of the drive train and detecting weak points in case operational exceedances or component fatigue. Airbus Helicopters undertakes research and developments for enhancing the reliability of Vibration Health Monitoring (VHM) using approaches for sensor data fusion and support vector data description in order to reduce the overall amount of required data for reliable machine state description.  

Session description
Speaker
Stefan BendischAirbus Helicopters
R&D Engineer
Airbus Helicopters

Multifunctional and high-precision sensor systems, modern network technologies and intelligent algorithms provide the foundation for innovative automated condition monitoring systems and predictive maintenance strategies for modern production facilities, decentralized devices and entire fleets. In his presentation Prof. Schulz will outline the importance of innovative data analysis and control software applying Artificial Intelligence as an essential part of the above-mentioned maintenance strategies. Severe damage events should be detected and avoided at an early stage.

Furthermore, existing use cases will be presented demonstrating the advantages such as minimised maintenance and, consequently, reduced operating costs and also an increased overall equipment effectiveness. The presentation will focus on wind turbines that are commonly controlled with SCADA systems and other sensory systems (e.g. on the basis of the machine sound analysis). The multi-stage processing of the large number of state information required for the clear identification of an emerging or pronounced damage pattern is explained using the example of this machine class.

Session description
Speaker
Prof. Dr. Michael SchulzIndalyz Monitoring & Prognostics (IM&P)
CEO
Indalyz Monitoring & Prognostics (IM&P)
Deep Dive:

Neural networks have been proven to be universal approximators but this still leaves the identification task a hard one. Beside using data we should focus our attention on the underlying structure of our subject of interest. In case of dynamical systems this is time, leading us to state space models and recurrent neural networks. After an introduction of small (open) dynamical systems you will study dynamical systems on manifolds. Here manifold and dynamics have to be identified in parallel. You will move on to large (closed) dynamical systems with hundreds of state variables and will combine causal and retro-causal models of the observations. This combination leads to an implicit description of dynamical systems on manifolds. Finally you will discuss the quantification of uncertainty in forecasting. In our framework the uncertainty appears as a consequence of principally unidentifiable hidden variables in the description of large systems. Together with the mathematical concepts you will see applications in economics and engineering.

Session description
Speaker
Dr. Hans-Georg ZimmermannFraunhofer Institut
Senior Research Scientist
Fraunhofer Gesellschaft
12:20 pm
Lunch Break
1:20 pm
Case Studies: Manufacturing, Energy & Health

HP Inc's Supplies Organization have been charting its own path towards Industry 4.0 since 2014. The biggest challenge is how to bring an organisation with 50+ production lines worldwide to Industry 4.0 performance seamlessly to achieve 20% productivity by 2020. This presentation will share how Manufacturing Analytics was developed and applied to our existing (& improving) data to reach industry 4.0 like performance. A case using predictive maintenance (without sensors) and predictive quality (to reduce destructive testing) will be shared.

Session description
Speaker
Richard LimHP
Manufacturing Strategist
HP

Today Industry 4.0 capabilities/technologies are progressing very fast and so are companies trying out new technologies. In this session, Amit will share the building blocks that were needed to get started with an Industry 4.0 program at Procter & Gamble India (technically and organisationally). He will present the latest industry trends and maturities and how the FMCG industry is looking at this problem as well as the major technological and innovation needs.

Session description
Speaker
Amit KurhekarProcter & Gamble
Senior Technology Manager
Procter & Gamble India
Deep Dive:

Deep Learning holds a strong promise for computer graphics from the topics of 3D rendering, content design generation to the creation of virtual worlds.The photo realistic image rendering is the process which turns artificial models into the larger than life images on the screen. Animators and game makers use costly rendering techniques called ray tracing to create high end images and videos. The high computational cost can be easily fathomed that one frame of a transformer character took average of 250 hours and we need 30 frames for 1 second of an animated movie. In this session, I will go through some of the break through work done to slash this rendering time drastically using deep learning with examples and codes. Moreover, We will explore with example codes and case studies the new frontiers of using advanced machine learning algorithms like Generative Modelling (GANs) , reinforcement learning and neural style transfer for graphics related tasks of Image denoising , image impainting and super resolution. The hacks to build real world AI based graphics application will also be covered.

Session description
Speaker
Muzahid HussainDassault Systems
R&D Engineer
Dassault Systems
2:20 pm
Short Break
2:25 pm
Case Studies: Manufacturing, Energy & Health

Data and the resulting sequence change from “algorithm -> data -> decision” towards “data -> algorithm -> decision“ drive a revolutionary change. Algorithms can help improve Overall Equipment Effectiveness, representing machine availability and performance as well as the quality of produced goods. The goal – improving OEE – is not new. New is the data-based approach by means of machine learning algorithms. An „edge“ solution close to or part of field devices and machines addresses the decision maker’s unwillingness moving production data into the cloud, allows data to stay in the production line and to be processed on the spot.

Session description
Speaker
Peter SeebergSofting Industrial
Business Development Manager
Softing Industrial

Our client – an industrial packaging company – wanted to calculate equipment efficiency and remotely monitor its production facilities. Normally, the answer would be to "throw more sensors at the problem" (vibration, sound, etc.). However, while adding sensors is cheap, using them is not (calibration, maintenance, etc.). So we looked at a different approach: What if we use the machine itself as our sensor, i.e., all we do is look at the electricity consumption to predict machine states? This case study will present our experience and challenges with this approach of “more analytics, less sensors”.

Session description
Speaker
Dr. Sebastian WernickeOne Logic
Chief Data Scientist
ONE LOGIC
Deep Dive:

This talk will focus on different methods that we can use to get clear insights from ML models. We see many data-products not going into production because it’s hard to justify and explain the predictions. It ‘s particulary true with factories where field users are reluctant to use black-box models when it involves machines and production lines. Interpretability is a major concern to onboard field users in data projects. We must be able to justify any choice we do with data. That’s why we will walk through several technics to tackle these interpretability challenges.

Session description
Speaker
Pierre PfennigDataiku
Senior Data Scientist
Dataiku
3:25 pm
Coffee Break
3:55 pm
Case Studies: Manufacturing, Energy & Health

The automation of business processes through predictive analytics is a challenge in both planning and realization. Bastian and Lukas will show how customer-value-oriented automation of business processes through predictive analytics can be implemented and integrated into the company-wide business processes. They have automated the processing and verification of meter readings at the Yello Strom GmbH for their plausibility. The integration into the corporate workflow was done using Microsoft Azure Machine Learning Services. In particular, they elaborated where the combination of human expertise with automation offers an advantage and thus create special added value.

Session description
Speakers
Bastian StockYello Strom GmbH
M.Sc. cand. Business Information Systems (Wirtschaftsinformatik)
Yello Strom
Lukas ZeisbergerYello Strom GmbH
M.Sc. cand. Business Information Systems (Wirtschaftsinformatik)
Yello Strom

Our client Evergreen Energy asked us to build a software that is able to predict from image data the number of solar panels which can be fit on a specific house. We developed an application which only needs to be fed with address data as input and delivers within seconds accurate estimates of the roof area and the number of solar panels. We will first give a live-demonstration of the application and then move on to explain the inner workings, especially the image segmentation part through the use of a neural net.

Session description
Speaker
Philipp JackmuthDida Datenschmiede
Founder
dida Datenschmiede GmbH
Deep Dive:
The Session Description will be available shortly.
Session description
4:55 pm
Short Break
5:00 pm
Closing Featured Session:

Progress in Generative Adversarial Networks. Let's close our eyes and imagine the world in two years. Companies such as NVIDIA will bring GAN technology to an industrial level. A trained GAN will than be able to create any image. Any image! On call, rapid, reliable, based on textual or graphical descriptions. This will have a revolutionary impact on photography and design related industries. These revolution is powered by a technology named transfer learning. Ulf will present a customer case in wich he used a GAN to predict the best mixture of color pigments for color production.

Session description
Speaker
Ulf SchönebergThe unbelievable Machine Company
Senior Data Scientist
The unbelievable Machine Company
5:30 pm
End of Predictive Analytics World for Industry 4.0 München 2018
CloseSelected Tags: