PAW for Industry 4.0 Berlin Agenda 2022

Berlin - October 5-6, 2022

International FMCG manufacturer Beiersdorf needs to forecast 1000s of products every month. In 2021, 10 years after the 1st Neural Networks were introduced, Beiersdorf set out to improve automatic forecasting further by reviewing the latest developments in technology. Surprisingly, some of the most recent and hyped algorithms such as DeepLearning, XGBoost, Prophet, BSTS and others did not perform well, but simple AI-methods customised to their data properties improved accuracy significantly.

A malt yield forecast with an excellent prediction performance, as well as first transfers to Augustiner Bräu, were successfully implemented to optimize the beer brewing process. The crucial next step is getting our ready-to-use analysis modules with built-in requirements into the running production. For this, we are creating an architecture for robust deployment, considering model resilience and automated detection of data drift and performance decay to eventually trigger new model training.

Global enterprises are increasingly relying on data analytics for decision making. To process data, firms leverage cloud-based data warehouses. As more on-prem data is moved to the cloud, the need for robust data governance controls to ensure data integrity, security, and regulatory adherence is mounting; however, existing governance processes are lagging. Here we present a zero-trust approach that can augment existing governance models and reduce exposure of sensitive data like PII.

The sound-based machine learning solution developed for Hindustan Shipping Limited helps identify defects that are sub surface or interior to the part. The identification of defect is real time, during the part production. Thus, enabling one to take actions immediately and not wait to produce a scrap part. Takeaways of this presenation are:(1) insights into the sound based machine learning approach for sub surface and interior defect detection; (2) how to identify the location and magnitude of defect in real time.

In recent years, there have been various efforts to product thinking and decentralized data loading using data meshes. However in deep learning, data loading is still challenging to master at scale. In this talk, we present our decentralized data loading solution and show why flexibility and collaboration is key to enable novel ML use cases. We hope to make large-scale model training accessible to a wider community and move towards more sustainable ML.

By predicting problems the laser machine availability can be increased significantly. This will not only reduce the costs of the maintenance. Started as a pure condition monitoring portal, the project for Trumpf Laser evolved into a hollistic predictive maintenance system, which allows facilitating the work of other departments (e.g. customer support).It also was the starting point for a new service: proactive support. Those practical examples shows the importance of empowering data-driven intelligence for machine manufacturers.

During this session, we will discuss the different risk-based categories of AI laid out by the EU’s Artificial Intelligence Act and find out how to become more admissible as per the Act. Thereafter, we will walk through the concrete steps, tools, and practices such as monitoring, explainability, model fairness, and compliance that are instrumental in achieving Responsible AI and building more risk-proof and market-friendly solutions.

Estimating lifetime of a machine or a wearing-part in a complex machine is a requirement in cost-optimal maintenance planning to reduce costly downtime or avoid too frequent maintenance. Regression models mapping features to the time-to-failure are not suited in the real world because of censored data. This talk shows how survival regressions can be utilized for a maintenance planning application. We discuss and demonstrate available survival analysis tools, their strengths and limitations.

Machine learning models are becoming obsolete and must be retrained – this is the current widespread tenor. Is this actually true? And if yes, which components does a CI/CD pipeline for machine learning really need – and which are optional? How can the whole thing be implemented without building a complete Machine Learning Platform team? And which challenges are still difficult to solve at present? A field report including (mis)decisions, which will help to choose the right path for your own challenges.

The desire to embody Responsible AI practices requires an understanding of, the context around, and the impact on the end user. To achieve this, design and research are just as pivotal to the RAI conversation as ML. There is no bigger risk, and no greater irresponsibility, than to not interface with those who will be affected by your design. This talk will share how to navigate customer relationships to encourage end user contact and mitigate assumptions and therefore risk.

In this session, Dr. Yvonne Blum will present how machine learning techniques in MATLAB were used to detect error conditions in MAN Energy Solutions gas turbines. These engines are distributed all over the world, quite often located in very remote areas where machine failure can have severe consequences. The goal of the project was to automate the time-consuming process of visualizing and manually evaluating measured sensor data, to determine gas turbine error conditions at an early stage. 

The case study was authored Dr. Holger Huitenga and Dr. Yvonne Blum.

Assembling a car is a complex task consisting of many steps usually grouped and organized in work plans. Based on the car model and its specifications, creating a key performance indicator (KPI) optimized work plan can be very time consuming. This case study at Volkswagen shows how data science can assist and speed up this process. After using various text analytics methods to identify similar work plans descriptions, a semi-automated benchmarking approach provides a KPI-driven recommendation.

Classification is one of the most frequently solved problems using machine learning. Unfortunately, it cannot handle a case with a number of classes, varying over time, and require all the data to be labelled. There is another approach, designed to solve cases when we can’t perform full data annotation and/or would like to dynamically modify the number of classes. Similarity learning is capable of solving such problems even with extreme classification. We’re going to show how to use such models in production.

Quality improvement for mass production lines is an ongoing topic for many years. The target is to reduce the rate of defects and scrap during production having an impact on sustainability, delivery time and cost. For the example of mass lens production at ZEISS we introduce a Markov-based method, that allows us to trace the movement of a given product through the production line to help us understand potential root causes for quality losses and thus being able to predict defects. In the end we aim to achieve a closed loop quality control avoiding quality losses by an improved understanding of root causes and proactive actions enabled by Industry 4.0 technologies such as Machine Connectivity and Artificial Intelligence.

AI algorithms deteriorate and fail silently over time impacting the business’ bottom line. The talk is focused on learning how you should be monitoring machine learning in production. It is a conceptual and informative talk addressed to data scientists & machine learning engineers. We’ll learn about the types of failures, how to detect and address them.

After quick successful POCs, the productive rollout of AI solutions often comes with unexpected challenges, especially for human-in-the-loop applications. The presentation will illustrate a holistic solution in three sections, starting with an end-to-end overview with the example of ML-based assistance systems, followed by deep-dives into the two main pain points: Imputation approaches for dealing with imperfect data and the psychology behind motivators for human interaction with such systems.

Fraud can be considerably reduced via speed, scalability, and stability. Investigating fraudulent activities, using fraud detection machine learning is crucial where decisions need to be made in microseconds, not seconds or even milliseconds. This becomes more challenging when things get demanding and scaling real-time fraud detection becomes a bottleneck. The talk will address these issues and provide solutions using the Hazelcast Open Source platform.

This table discussion focuses on how data scientists, data nerds, head of data science, and chief data officers can contribute to sustainability in AI. We would like to discuss this from different views like science contribution to SDGs as well as running climate-neutral model developments. The participants should get some understanding on the carbon footprint of doing intensive modeling and what the industry can do to reduce it.

The changes in the ads ecosystem have led marketers to lean on existing aggregate experimentation tools that assume a predetermined treatment effect. Choosing the treatment group to ensure you have high chances of detecting an effect is non-trivial. Built by Meta Open Source, GeoLift solves this problem by building well powered geographical experiments. Join us to go over why geographical experiments are necessary and their implications in the marketing industry, along with a demo of GeoLift.

Telefónica, the biggest Spanish telecommunications company, asked us to provide a machine learning solution capable of detecting when one of their DSLAMs has an anomaly in registered customers indicating a loss in customer IPTV service. You will be shown how to deal with thousands of time series data by combining clustering algorithms, smoothing methods and deep learning tools to obtain efficient and high-performance results.