Daniel Weimer

Machine learning in manufacturing: advantages, challenges, and applications

The nature of manufacturing systems faces ever more complex, dynamic and at times even chaotic behaviors. In order to being able to satisfy the demand for high-quality products in an efficient manner, it is essential to utilize all means available. One area, which saw fast pace developments in terms of not only promising results but also usability, is machine learning. Promising an answer to many of the old and new challenges of manufacturing, machine learning is widely discussed by researchers and practitioners alike. However, the field is very broad and even confusing which presents a challenge and a barrier hindering wide application. Here, this paper contributes in presenting an overview of available machine learning techniques and structuring this rather complicated area. A special focus is laid on the potential benefit, and examples of successful applications in a manufacturing environment.

Network structures and decentralized control in logistics: topology, interfaces, and dynamics

Decentralized or autonomous control in logistics has continuously developed over the last decades. Trends such as Ubiquitous Computing, Industry 4.0, and Cyber–physical Systems foster the transition from centralized control to decentralized approaches. While there has been a lot of research on control algorithms and miniaturization of information and communication technology, comparably little is known about the effect of the underlying structures of the logistic networks on decentralized control. This article aims at revisiting the topic of decentralized control with regard to the topology of logistic networks and at highlighting the importance of the relation between topological structure and control procedures in order to propose directions for future research.

A 3D-robot vision system for automatic unloading of containers

Unloading of standard containers within logistic processes is mainly performed manually. Amongst gripping technology, the development of a robot vision system for recognizing different shaped logistic goods is a major technical obstacle for developing robotic systems for automatic unloading of containers. Goods can be arbitrarily placed inside a container and the resulting packaging scenarios usually have a high degree of occlusion. Existing systems and approaches use range information acquired by laser scanners for recognizing and localizing goods inside of containers. They are restricted to a single shape class of goods and often have limited size ranges for goods. This paper presents a robot vision for recognizing and localizing differently shaped and sized objects in piled packaging scenarios using range data acquired by different kinds of range sensors. After a specific segmentation step, different shaped partial surfaces are detected and classified in point cloud data and combined to complete logistic goods. The system is evaluated with real and simulated sensor data from different packaging scenarios.

3D-Computer Vision for Automation of Logistic Processes

The availability of low-cost range sensors has led to several innovative implementations and solutions in various application fields like object recognition and localization, scene understanding, human-robot interaction or measurement of objects. The transfer of the corresponding methods and techniques to logistic processes needs the consideration of specific requirements. A logistic application field that requires robust and reliable 3D vision systems is automated handling of universal logistic goods for (de-)palletizing or unloading of standard containers in the field of sea and air cargo. This paper presents a 3D-computer vision system for recognizing and localizing different shaped logistic goods for automated handling by robotic systems. The objective is to distinguish between different types of goods like boxes, barrels or sacks due to their geometric shape in point cloud data. The system is evaluated with sensor data from a low-cost range sensor and ideal simulated data representing different shaped logistic goods as well.

Combined Categorization and Localization of Logistic Goods Using Superquadrics

The detection and pose estimation of various shaped objects in real world cluttered application scenarios is a major technical challenge due to noisy sensor data and possible occlusions. Usually, a predefined model database is utilized for implementing a robust and reliable object detection system. Geometric models based on superquadrics have shown great potential and flexibility for representing a variety of shapes by using only a few parameters. In this paper, we propose a novel method concerning superquadric based Segment-then-fit approach and evaluate it in a logistic application scenario. The method utilizes boundary and region information to recover different types of convex shaped logistic goods in cluttered scenarios for automated handling by means of unorganized point cloud data. We have evaluated our approach using synthetic and multiple real sensor data on several packaging scenarios with various shaped logistic goods.

3D OBJECT CATEGORIZATION OF LOGISTIC GOODS FOR AUTOMATED HANDLING

The automated handling of universal logistic goods through robotic systems requires suitable and reliable methods for categorizing different logistic goods. They must be able to detect the pose of different types and sizes of logistic goods in order to identify possible gripping points or for selecting a suitable gripping system for the detected object type. For this purpose, Time-of-Flight or Structured Light sensors can deliver a dense 3D representation of the investigated scenario. This paper presents a 3D object categorization system for logistic goods based on synthetically generated model data. We generate the model data by using a sensor simulation framework for different TOF-sensor types. The framework creates point clouds of self-defined geometric models of logistic goods or CAD data. Afterwards, we use these synthetic point clouds for generating a suitable model database offline. In order to evaluate our approach, we describe the synthetic point clouds by global point feature description techniques to distinguish between different types of logistic goods. Finally, we evaluate our concept with real sensor data from different logistic goods.