Visual analysis of blow molding machine multivariate time series data

J Vis https://doi.org/10.1007/s12650-022-00857-4 R E G UL A R P A P E R Maath Musleh • Angelos Chatzimparmpas • Ilir Jusufi Visual analysis of blow molding machine multivariate time series data Received: 1 February 2022 / Revised: 25 April 2022 / Accepted: 1 June 2022  The Author(s) 2022 Abstract The recent development in the data analytics field provides a boost in production for modern industries. Small-sized factories intend to take full advantage of the data collected by sensors used in their machinery. The ultimate goal is to minimize cost and maximize quality, resulting in an increase in profit. In collaboration with domain experts, we implemented a data visualization tool to enable decision-makers in a plastic factory to improve their production process. The tool is an interactive dashboard with multiple coordinated views supporting the exploration from both local and global perspectives. In summary, we investigate three different aspects: methods for preprocessing multivariate time series data, clustering approaches for the already refined data, and visualization techniques that aid domain experts in gaining insights into the different stages of the production process. Here we present our ongoing results grounded in a human-centered development process. We adopt a formative evaluation approach to continuously upgrade our dashboard design that eventually meets partners’ requirements and follows the best practices within the field. We also conducted a case study with a domain expert to validate the potential application of the tool in the real-life context. Finally, we assessed the usability and usefulness of the tool with a two-layer summative evaluation that showed encouraging results. Keywords Time series data  Unsupervised machine learning  Visualization 1 Introduction Modern production lines accommodate a high number of sensors and actuators with the aim of improving the quality and reliability of products, enhancing the efficiency of maintenance routines, and ensuring the proper working conditions for several machines (Glebke et al. 2019). These devices generate a wide variety of data ranging from environmental data (e.g., temperature and weather conditions) to data about the production process and the production state (e.g., if a machine is on or off). Analyzing historical data originating from the aforementioned sources can help to accurately forecast whether there will be a change in production speed due to external reasons (Tao et al. 2018). M. Musleh (&) Institute of Visual Computing and Human-Centered Technology, TU Wien, 1040 Vienna, Austria E-mail: A. Chatzimparmpas  I. Jusufi Department of Computer Science and Media Technology, Linnaeus University, Växjö 351 95, Sweden E-mail: I. Jusufi E-mail: M. Musleh et al. Fig. 1 Visualization of machine cycles with the multiple coordinated views of our tool. (A) The line plot presents the original values of different features in time series of the production cycle. B–D panel for the exploration of distinct data subsets with alternative methods. (E) The message banner to track the user’s selections. (F) The Heatmap displays the normalized values of the DTW-processed time series features. (G) The feature selection panel for choosing specific features to plot. (H) The useradjustable t-SNE’s hyperparameters. I–J The t-SNE and PCA plots form groups of points that can be examined further. (K) The PCP highlights the correlation between features. The colors used in I–K views are computed by applying k-means clustering to the PCA plot Smart factories rely on careful data management and the choice of appropriate tools for the analysis of data (Park et al. 2016). However, collecting and examining the data using statistical or computational approaches do not satisfy the requirements of an agile production environment (Gao et al. 2020). Domain expertise is crucial for data (and model) interpretation (Chatzimparmpas et al. 2020a) as each production line (or even machine) produces different sensor readings depending on many variables-for example, the air pressure at specific parts of a product and the type of the product. Therefore using visual analytics (VA) to involve the domain experts in the data analysis phase is inevitable (Chatzimparmpas et al. 2020b). The plastic industry employs several data-driven technologies to optimize the production process (Esposito 2019). Smart factories today produce plastic parts more efficiently and cost-effectively than ever before. Blow molding is one of the most common methods in the manufacturing of plastic (Altarazi et al. 2019). For simplicity, this process could be summed up in two stages: melting the parison (i.e., plastic material) and shaping it in the mold by an air blow (Yu and Juang 2010). Our industry partner develops robotic and machine learning (ML) solutions for such smart factories. One of their clients produces automobiles plastic parts using the blow molding process. Lately, they introduced a product that collects data remotely through the sensors installed in a blow molding machine. The accumulated data provide a potential for valuable insights related to the production process that can be used to assist factory’s management in decision-making. However, the exploration of multivariate time series data captured by the sensors poses a challenge (Zhou et al. 2018). Indeed, temporal data requires thoughtful methods to extract the important features from tightly coupled multidimensional data. Finally, settling on the optimal visualization techniques requires an extensive investigation of users’ prior experience and their needs (Bernard et al. 2019). Visual analysis of blow molding machine multivariate... Based on continuous discussions with our industry partner, we collected the following requirements (R1–R3): • (R1) highlight emerging patterns of the production process; • (R2) enable the identification of the important features that heavily affect the production process; and • (R3) cover any remaining gaps of their other deployed ML tools to facilitate even further data-driven decision-making. To satisfy the previously defined requirements, we present visualization techniques for interactive data analysis of remotely collected data in a plastic production factory. We preprocess the data and allow users to explore particular features using clustering and dimensionality reduction (DR) methods (Sacha et al. 2017), as shown in Fig. 1. Our proposed VA tool enables the factory’s management and technicians to make informed maintenance and production decisions. We follow a user-centric approach by involving a visualization expert and an industry partner from the early stages of development to improve our tool’s design iteratively. Overall, our contributions consist of the following: 1. A data visualization tool that comprises several visual representations arranged in a single webpage in order to sufficiently explore multivariate time series data from a (...truncated)