Batch runs corresponding to the same recipe usually have different duration. The data collected by the sensors that equip batch production lines reflects this fact: time series with different lengths and unsynchronized events. Dynamic Time Warping (DTW) is an algorithm successfully used, in batch monitoring too, to synchronize and map to a standard time axis two series, an action called alignment. The online alignment of running batches, although interesting, gives no information on the remaining time frame of the batch, such as its total runtime, or time-to-end. We notice that this problem is similar to the one addressed by Survival Analysis (SA), a statistical technique of standard use in clinical studies to model time-to-event data. Machine Learning (ML) algorithms adapted to survival data exist, with increased predictive performance with respect to classical formulations. We apply a SA-ML-based system to the problem of predicting the time-to-end of a running batch, and show a new application of DTW. The information returned by openended DTW can be used to select relevant data samples for the SA-ML system, without negatively affecting the predictive performance and decreasing the computational cost with respect to the same SA-ML system that uses all the data available. We tested the system on a real-world dataset coming from a chemical plant.
The manufacturing industry and, for this research, the automotive manufacturing industry specifically, is always on the lookout for opportunities to improve production throughput with a minimum of investment. Identifying these opportunities often requires the observation of the current production process by experts. This paper is the continuation of the previous work ’Automated, Nomenclature Based Data Point Selection for Industrial Event Log Generation’. One of its aims is to provide strategies that can be used to pre-process an in-depth, slightly flawed industrial equipment log to allow for further analysis. The pre-processing is achieved by identifying the flaws, removing the non-value added events and a heuristic methodology to cluster the log into individual sequences. Expert knowledge then is encoded into engineering features to extend the log matrix and prepare it for machine learning model generation for identification of the complete cases. To derive value from the available data, the sequences are plotted into Gantt charts, and eight hypotheses are introduced that allow for automated annotations within this chart to highlight potential areas of improvement. Application of the framework to real life logs, obtained from stations considered bottlenecks within the evaluated automotive body shop, lead to the discovery of improvement potential between two and twelve seconds per cycle.
The rise of artificial intelligence (AI) promises productivity gains in industrial practice. While IT technology offers a variety of technological advances, plant owners strive for stability and robustness of the production process. To overcome this tension field, we propose a set of 16 requirements for the development of industrial AI solutions to foster i) the adaptation process, ii) support the solution engineering and iii) ease the embedding into the existing system landscape while respecting iv) safety aspects to build up v) trust into industrial AI solutions. The proposed requirements can guide industrial stakeholders to focus on the right solution approach for specific production challenges and support them in voicing their own needs towards novel AI solutions. This will help AI developers to speed up time-to-market as well as to increase market acceptance of industrial AI solutions. Overall, specifying requirements on industrial AI will foster the acceptance and utilization rates of AI solutions in industrial practice.
Maintaining modern production machinery requires a significant amount of time and money. Still, plants suffer from expensive production stops and downtime due to faults within individual components. Often, plants are too complex and generate too much data to make manual analysis and diagnosis feasible. Instead, faults often occur unnoticed, resulting in a production stop. It is then the task of highly-skilled engineers to recognise and analyse symptoms and devise a diagnosis. Modern algorithms are more effective and help to detect and isolate faults faster and more precise, thus leading to increased plant availability and lower operating costs. In this paper we attempt to solve some of the described challenges. We describe a concept for an automated framework for hybrid cyberphysical production systems performing two distinct tasks: 1) fault diagnosis and 2) reconfiguration. For diagnosis, the inputs are connection and behaviour models of the components contained within the system and a model describing their causal dependencies. From this information the framework is able to automatically derive a diagnosis provided a set of known symptoms. Taking the output of the diagnosis as a foundation, the reconfiguration part generates a new configuration, which, if applicable, automatically recovers the plant from its faulty state and resumes production. The described concept is based on predicate logic, specifically Satisfiability-Modulo-Theory. The input models are transformed into logical predicates. These predicates are the input to an implementation of Reiter’s diagnosis algorithm, which identifies the minimum-cardinality diagnosis. Taking this diagnosis, a reconfiguration algorithm determines a possible, alternative control, if existing. Therefore the current system structure described by the connection and component models is analysed and alternative production plans are searched. If such an alternative plan exists, it is transmitted to the control of the system. Otherwise, an error that the system is not reconfigurable is returned.
In this paper, a new information model for machine learning applications is introduced, which allows for a consistent acquisition and semantic annotation of process data, structural information and domain knowledge from industrial productions systems. The proposed information model is based on Industry 4.0 components and IEC 61360 component descriptions. To model sensor data, components of the OGC SensorThings model such as data streams and observations have been incorporated in this approach. Machine learning models can be integrated into the information model in terms of existing model serving frameworks like PMML or Tensorflowgraph. Based on the proposed information model, a tool chain for automatic knowledge extraction is introduced and the automatic classification of unstructured text is investigated as a particular application case for the proposed tool chain.
Long underwater operations with autonomous battery charging and data transmission require an Autonomous Underwater Vehicle (AUV) with docking capability, which in turn presume the detection and localization of the docking station. Object detection and localization in sonar images is a very difficult task due to acoustic image problems such as, non-homogeneous resolution, non-uniform intensity, speckle noise, acoustic shadowing, acoustic reverberation and multipath problems. As for detection methods which are invariant to rotations, scale and shifts, the Generalized Fuzzy Hough Transform (GFHT) has proven to be a very powerful tool for arbitrary template detection in a noisy, blurred or even a distorted image, but it is associated with a practical drawback in computation time due to sliding window approach, especially if rotation and scaling invariance is taken into account. In this paper we use the fact that the docking station is made out of aluminum profiles which can easily be isolated using segmentation and classified by a Support Vector Machine (SVM) to enable selective search for the GFHT. After identification of the profile locations, GFHT is applied selectively at these locations for template matching producing the heading and position of the docking station. Further, this paper describes in detail the experiments that validate the methodology.
To demonstrate how deep learning can be applied to industrial applications with limited training data, deep learning methodologies are used in three different applications. In this paper, we perform unsupervised deep learning utilizing variational autoencoders and demonstrate that federated learning is a communication efficient concept for machine learning that protects data privacy. As an example, variational autoencoders are utilized to cluster and visualize data from a microelectromechanical systems foundry. Federated learning is used in a predictive maintenance scenario using the C-MAPSS dataset.
Information processing systems with some form of machine-learned component are making their way into the industrial application and offer high potentials for increasing productivity and machine utilization. However, the systematic engineering approach to integrate and manage these machine-learned components is still not standardized and no reference architecture exist. In this paper we will present the building block of such an architecture which is developed with the ML4P project by Fraunhofer IFF.
Reconstruction of highly porous structures from FIB-SEM image stacks is a difficult segmentation task. Supervised machine learning approaches demand large amounts of labeled data for training, that are hard to get in this case. A way to circumvent this problem is to train on simulated images. Here, we report on segmentation results derived by training a convolutional neural network solely on simulated FIB-SEM image stacks of realizations of a variety of stochastic geometry models.
Condition monitoring systems based on deep neural networks are used for system failure detection in cyber-physical production systems. However, deep neural networks are vulnerable to attacks with adversarial examples. Adversarial examples are manipulated inputs, e.g. sensor signals, are able to mislead a deep neural network into misclassification. A consequence of such an attack may be the manipulation of the physical production process of a cyber-physical production system without being recognized by the condition monitoring system. This can result in a serious threat for production systems and employees. This work introduces an approach named CyberProtect to prevent misclassification caused by adversarial example attacks. The approach generates adversarial examples for retraining a deep neural network which results in a hardened variant of the deep neural network. The hardened deep neural network sustains a significant better classification rate (82
The prediction of the power consumption increases the transparency and the understanding of a cutting process, this delivers various potentials. Beside the planning and optimization of manufacturing processes, there are application areas in different kinds of deviation detection and condition monitoring. Due to the complicated stochastic processes during the cutting processes, analytical approaches quickly reach their limits. Since the 1980s, approaches for predicting the time or energy consumption use empirical models. Nevertheless, most of the existing models regard only static snapshots and are not able to picture the dynamic load fluctuations during the entire milling process. This paper describes a data-driven way for a more detailed prediction of the power consumption for a milling process using Machine Learning techniques. To increase the accuracy we used separate models and machine learning algorithms for different operations of the milling machine to predict the required time and energy. The merger of the individual models allows finally the accurate forecast of the load profile of the milling process for a specific machine tool. The following method introduces the whole pipeline from the data acquisition, over the preprocessing and the model building to the validation.
Machine learning and deep learning are widely used in various applications to assist or even replace human reasoning. For instance, a machine learning based intrusion detection system (IDS) monitors a network for malicious activity or specific policy violations. We propose that IDSs should attach a sufficiently understandable report to each alert to allow the operator to review them more efficiently. This work aims at complementing an IDS by means of a framework to create explanations. The explanations support the human operator in understanding alerts and reveal potential false positives. The focus lies on counterfactual instances and explanations based on locally faithful decision-boundaries.
Against the backdrop of the economically and ecologically optimal management of electrical energy systems, accurate predictions of consumption load profiles play an important role. On this basis, it is possible to plan and implement the use of controllable energy generation and storage systems as well as energy procurement with the required lead-time, taking into account the technical and contractual boundary conditions. The recorded electrical load profiles will increase considerably in the course of the digitization of the energy industry. In order to make the most accurate predictions possible, it is necessary to develop and investigate models that take account of the growing quantity structure and, due to the significantly higher number of observations, improve the forecasting quality as far as possible. Artificial neural networks (ANN) are increasingly being used to solve non-linear problems for a growing amount of data that is affected by human and other unpredictable influences. Consequently, the model approach of an ANN is chosen for predicting load profiles. Aim of the thesis is the simulative investigation and the evaluation of the quality and optimality of a prediction model based on an ANN for electrical load profiles.
In this contribution, we give an insight in our experiences in the technical and organizational realization of industrial analytics. We address challenges in implementing industrial analytics in real-world applications and discuss aspects to consider when designing a machine learning solution for production. We focus on technical and organizational aspects to make industrial analytics work for real-world applications in factory automation. As an example, we consider a machine learning use case in the area of industry compressors. We discuss the importance of scalability and reusability of data analytics pipelines and present a container-based system architecture.
This paper shows how automation components can be enhanced with self-monitoring capabilities, which are more effective than traditional rule-based methods, by using Industrial Analytics approaches. Two application examples are presented to show how this approach allows the realization of a predictive maintenance strategy, while drastically reducing the realization effort. Furthermore, the benefits of a flexible architecture combining edge- and cloud-computing for the realization of such monitoring system are discussed.
Today, the optimization of the press hardening process is still a complex and challenging task. This report describes the combination of linear regression with least squares optimization to adjust the process parameters of this process for quality improvement. The FE simulation program AutoForm was used to model the production line concerned and various process and quality parameters were measured. The proposed system is capable of automatically adjusting the process parameters of following process steps based on the quality estimate at each step of the production line. An additional benefit is the identification of likely defective parts early in the production process. Based on the results derived from 1000 observations a better understanding of the process was obtained and in the future the combined regression and optimization approach can be extended to more complex production lines.
Cyber Physical Production Systems (CPPS) provide a huge amount and variety of process and production data. Simultaneously, operational decisions are getting ever more complex due to smaller batch sizes (down to batch size one), a larger product variety and complex processes in production systems. Production engineers struggle to utilize the recorded data to optimize production processes effectively. In contrast, CPPS promote decentralized decision-making, so-called intelligent agents that are able to gather data (via sensors), process these data, possibly in combination with other information via a connection to and exchange with others, and finally take decisions into action (via actors). Modular and decentralized decision-making systems are thereby able to handle far more complex systems than rigid and static architectures. This paper discusses possible applications of Machine Learning (ML) algorithms, in particular Reinforcement Learning (RL), and the potentials towards an production planning and control aiming for operational excellence.
Small humps on the floor go beyond the detectable scope of laser scanners and are therefore not integrated into SLAM based maps of mobile robots. However, even such small irregularities can have a tremendous effect on the robot's stability and the path quality. As a basis to develop anomaly detection algorithms, kinematics data is collected exemplarily for an overrun of a cable channel and a bulb plate. A recurrent neuronal network (RNN), based on the autoencoder principle, could be trained successfully with this data. The described RNN architecture looks promising to be used for realtime anomaly detection and also to quantify path quality.
With an ongoing digital transformation towards industry 4.0 and the corresponding growth of collected sensor data based on cyber-physical systems, the need for automatic data analysis in industrial production lines has increased drastically. One relevant application scenario is the usage of intelligent approaches to anticipate upcoming failures for maintenance. In this paper, we present a novel approach for anomaly detection regarding predictive maintenance in an industrial data-intensive environment. In particular, we are focusing on historical sensor data from a real reflow oven that is used for soldering surface mount electronic components to printed circuit boards. The sensor data, which is provided within the scope of the EU-Project COMPOSITION (under grant no. 723145), comprises information about the heat and the power consumption of individual fans inside a reflow oven. The data set contains time-annotated sensor measurements in combination with additional process information over a period of more than seven years.
This paper presents an application of a random forest based classifier that aims at recognizing flawed products in a highly automated production environment. Within the course of this paper, some data set and application features are highlighted that make the underlying classification problem rather complex and hinders the usage of machine learning algorithms straight out-of-the-box. The findings regarding these features and how to treat the concluded challenges are highlighted in a abstracted and generalized manner.
