It’s important to make sure all analytic work is reviewed properly to avoid deriving any misleading conclusions. Consider this hypothetical: You work for a company developing and manufacturing medical devices. The COVID-19 pandemic created a worldwide shortage of ventilators. Since your company has recently implemented a major digital transformation strategy to meet Industry 4.0 standards, you are able to predict operating failures in alternative assembly lines, provide online monitoring of wave soldering processes, and gather focused statistics on assembly defects from automated visual inspection robots. The flexibility acquired by this digital transformation permitted the rapid conversion of the company’s production lines to make the much-needed mechanical ventilators. A major element in this transformation is the application of analytics, since the flexibility described above requires high-level analytic capabilities. Keep reading for additional background and a checklist for reviewing analytic-based reports. Industry 4.0, the so-called fourth industrial revolution, relies on three basic elements: Sensor technology that can extensively measure products and processes online. Flexible manufacturing capabilities – such as 3D printing – that can efficiently produce batches of varying size. Analytics that power the industrial engine with the capability to monitor, diagnose, predict and optimize decisions. One significant analytic challenge is data integration, since sensors may collect data in different time cycles. Dynamic time warping (DTW) or other techniques may be the best way to get a holistic view. Most likely, any analytic work done in industry means that the data has been collected actively or passively and that models have been developed with empirical methods, first principles or hybrid models. The industrial cycle provides short-term opportunities to try out new products or new process set-ups and, based on the results, determine follow-up actions. But it’s important to make sure all analytic work is reviewed properly to avoid deriving any misleading conclusions, which could be very costly and/or time-consuming. For example, a lithium battery manufacturer discovered it had uncalibrated test equipment evaluating end-of-the-line products. The company was able to avoid a major recall by using the plant’s control charts to precisely identify the problematic batches. To avoid shipping immature products or defective batches, good diagnostic capabilities are vital for identifying the cause of any reported problems. Specific analytic challenges in systems engineering and industrial applications include: Engineering design Manufacturing systems Decision-support systems Shop-floor control and layout Fault detection and quality improvement Condition-based maintenance Customer and supplier relationship management Energy and infrastructure management Cybersecurity and security  These challenges generate projects for better monitoring products and process; designing new or converting existing products and processes; and improving products and processes. To address these problems by using data, proper access to analytic methods is necessary. One approach to review analytic work is to use the information quality framework that I presented at JMP Discovery Summit Prague. Information quality is determined by the goal of the analysis, the methods of analysis used in the study, the data resolution with respect to the study goal, data integration, and communicating the findings. (For more on this topic, including how to set up an information quality evaluation workshop, you may be interested in The Real Work of Data Science.) Information quality (InfoQ), a concept Galit Shmueli and I introduced in Information Quality: The Potential of Data and Analytics to Generate Knowledge, is a general framework for planning, tracking and assessing the value of data analytics. InfoQ is based on four components and eight dimensions. It is defined as “the utility of a particular data set for achieving a given analysis goal by employing statistical analysis or data mining.” InfoQ is affected by the analysis goal (g), the available data (X), the implemented data analysis (f) and study utility (U). In evaluating an analytic project, each of these four components and the eight information quality dimensions must be examined. The eight InfoQ dimensions are:  Data resolution Data structure Data integration Temporal relevance Generalizability Chronology of data and goal Operationalization Communication Data resolution needs to match the study goal so the outcomes of the analysis must be provided at the right time to the right decision maker. For example, some image processing analytics used to evaluate quality of a fast-running paper production process must produce timely results so that the quality engineer can act properly. If a precise imaging classifier, used to identify quality issues hidden in a paper roll, is delayed because of computational constraints, the information it produces will be of very low quality. The checklist below can be used to assess the eight information quality dimensions of a specific study. A JMP add-in for summarizing the information quality assessment is available in the Community. Examples of how to use JMP to generate information quality can be downloaded here. The Checklist These are the eight questions to ask when reviewing an analytic study after clarifying the goals and utility. Dimension Questions Data resolution Is the data granularity adequate for the intended job? Has measurement uncertainty been evaluated and found appropriate? Data structure Is it possible to use data from different sources that reflect on the problem at hand? Data integration How is data from different sources integrated? Are there linkage issues that lead to dropping crucial information? Temporal relevance Does the time gap between data collection and analysis cause any concern? Chronology of data and goal Are the analytic findings communicated to the right persons in a timely manner? Generalizability Can general conclusions be derived beyond what was explicitly studied? For example, conclusions that can be applied to other products or processes. Operationalization Are the measured variables themselves relevant to the study goal? Are there any stated action item recommendations derived from the study? Communication Are findings properly communicated to the intended audience?   These questions should help the reviewers who need to assess an analytics-based study in Industry 4.0 applications and beyond. However, they are not meant to be prescriptive and will need to be adapted to specific situations.  In 2020, the Fourth Industrial Revolution continues to grow in advanced manufacturing, combining extensive sensor data with flexible manufacturing and advanced analytics. A new comprehensive book, covering state-of-the-art elements of system engineering and the Fourth Industrial Revolution, is  available here.   For More Information Learn more through these JMP resources related to Industry 4.0: A Practical Road Map for Implementing Industry 4.0 Using Custom SQL Databases and JMP 15 Joseph Beauchemin and Philip Ramsey Why JMP Pro is the Analytics Tool of Choice for Industry 4.0 Bernard McKeown and Phil Kay What is the best improvement strategy for Industry 4.0? Anne Milley Production Line Control in Semiconductor High-Volume Manufacturing Guillaume Bugnon and Stéphane Astie  Information quality in the golden age of analytics Ron Kenett  Calculate 'InfoQ' score with JMP Ian Cox  ENBIS COVID-19 WEBINAR SERIES: Analytics in Industry 4.0 with some Covid-19 Implications Ron Kenett  K-means clustering analysis of seven sensors tracking possible deformation under stress, shown in the ENBIS webinar listed above. ... View more

QbD update: a look back and to the future to see how QbD can help pharma and other industries streamline and optimize processes for state-of-the-art results. Quality by Design (QbD) is about processes, products and clinical understanding. Five years ago, my colleagues and I launched a QbD column , which consisted of nine entries. I encourage you to review the original posts in the QbD column since they are still very relevant today. Before you revisit the column, take a moment to watch this four-minute video of the Sanofi factory of the future to see an advanced implementation of biotechnology, chemistry, manufacturing and controls (CMC). To appropriately manufacture a drug product, a specific manufacturing process, product characteristics, and product testing must all be defined to ensure that the product is safe, effective and consistent between batches. The QbD column series provides a peek under the hood on how the pharmaceutical industry develops and optimizes products and processes, so that they meet the requirements of the factory of the future. Let me give you an overview of each column: Overview of QbD, which covers five sections: What is QbD? How does QbD improve quality? Design space in QbD Role of experiments in QbD QbD and drug development These are all very brief sections designed to set up the context for the following columns. A factorial design case study based on the formulation of a steroid lotion of a generic product. The design was set to optimize eight quality attributes simultaneously by setting temperature, blending time and cooling time. This study shows how JMP simulates responses to visualize how the variability in factor-level settings affects the variability in responses. A QbD fractional factorial experiment. The goal of the experiment is to explore the process of preparing nanosuspensions, a popular formulation for water-insoluble drugs. Nanosuspensions involve colloidal dispersions of discrete drug particles, which are stabilized with polymers and/or surfactants. The experiment involves four three-level factors and one two-level factor. Achieving robustness with stochastic emulators. Stochastic emulators were first proposed in Bates et al, 2006 to achieve robust, on-target performance. This very powerful approach is implemented in JMP. The method is demonstrated with a case study that refers to the standard on Scale-Up and Post-Approval Changes (SUPAC) of the FDA Center for Drug Evaluation and Research (CDER). Mixture designs . For these designs, the emphasis is on the effect of the relative proportions of ingredients in a formulation, rather than their absolute amounts. The column shows examples of ternary plots and profilers tailored to treat mixtures. The example used is a case study on mixing powders. Key questions for the blending operation are: How to quantify components of powder blends simultaneously? How to validate or confirm the process analytical technology (PAT) blending process monitoring results via other fast and convenient spectroscopic methods? How to link the scale of scrutiny and the homogeneity of both API and excipients? Response surfaces and sequential experimentation. In this post, a Placket Burman (PB) design followed by a central composite design (CCD) is featured in a case study that considers three immediate outcomes from an experiment on formulations: encapsulation efficiency (EE), particle size and zeta potential. In addition, each formulation was diluted and divided into samples for storage, half at 4⁰ C and half at 37⁰ C. These samples were tested at predetermined times (up to 24 months) to detect loss of active drug during storage. Split plot experiments. These experiments have hard-to-change factors that are difficult to randomize and can only be applied at the block level. Once the level of a hard-to-change factor is set, experiments are run with several other factors that keep that level fixed. The case study used in this column, which is from preclinical research on animal models where several methods for the treatment of severe chronic skin irritations are tested, demonstrates how to design and analyze such experiments. Is QbD applicable for developing analytic methods? This post is based on a paper I wrote with my brother, who was at Teva at the time (Kenett and Kenett, 2008). The case study is from the development of a High Performance Liquid Chromatography (HPLC) method.  Applying QbD to make analytic methods robust. This post shows how definitive screening designs (DSD) can be used to assess and improve analytic methods. The material and examples contained in the QbD columns are still relevant. Some additional developments account for the growing impact of sensor technology, big data, machine learning and hybrid models. Eventually, the challenge is to generate information from all the data gathered in the laboratory and the factory floor; an even greater challenge is to ensure the quality of this information, which I discuss in this conversation with Anne Milley (a transcript is available here). A video of a more formal presentation at a JMP QbD discovery event in Copenhagen is available here. In 2020, the Fourth Industrial Revolution continues to grow in advanced manufacturing, combining extensive sensor data with flexible manufacturing and advanced analytics. A new comprehensive book, covering state-of-the-art elements of system engineering and the Fourth Industrial Revolution, is available here.  I spoke with Anne Milley about "Quality Assurance in the Golden Age of Analytics" for the Analytically Speaking video series.   ... View more

https://www.linkedin.com/pulse/ten-questions-statistics-data-science-2020-beyond-ron-s-kenett/ ... View more