David Lechevalier

A neural network meta-model and its application for manufacturing

Manufacturing generates a vast amount of data both from operations and simulation. Extracting appropriate information from this data can provide insights to increase a manufacturers competitive advantage through improved sustainability, productivity, and flexibility of their operations. Manufacturers, as well as other industries, have successfully applied a promising statistical learning technique, called neural networks (NNs), to extract meaningful information from large data sets, so called big data. However, the application of NN to manufacturing problems remains limited because it involves the specialized skills of a data scientist. This paper introduces an approach to automate the application of analytical models to manufacturing problems. We present an NN meta-model (MM), which defines a set of concepts, rules, and constraints to represent NNs. An NN model can be automatically generated and manipulated based on the specifications of the NN MM. In addition, we present an algorithm to generate a predictive model from an NN and available data. The predictive model is represented in either Predictive Model Markup Language (PMML) or Portable Format for Analytics (PFA). Then we illustrate the approach in the context of a specific manufacturing system. Finally, we identify future steps planned towards later implementation of the proposed approach.

Standards based generation of a virtual factory model

Developing manufacturing simulation models usually requires experts with knowledge of multiple areas including manufacturing, modeling, and simulation software. The expertise requirements increase for virtual factory models that include representations of manufacturing at multiple resolution levels. This paper reports on an initial effort to automatically generate virtual factory models using manufacturing configuration data in standard formats as the primary input. The execution of the virtual factory generates time series data in standard formats mimicking a real factory. Steps are described for auto-generation of model components in a software environment primarily oriented for model development via a graphic user interface. Advantages and limitations of the approach and the software environment used are discussed. The paper concludes with a discussion of challenges in verification and validation of the virtual factory prototype model with its multiple hierarchical models and future directions.

Automated uncertainty quantification analysis using a system model and data

Understanding the sources of, and quantifying the magnitude of, uncertainty can improve decision-making and, thereby, make manufacturing systems more efficient. Achieving this goal requires knowledge in two separate domains: data science and manufacturing. In this paper, we focus on quantifying uncertainty, usually called uncertainty quantification (UQ). More specifically, we propose a methodology to perform UQ automatically using Bayesian networks (BN) constructed from three types of sources: a descriptive system model, physics-based mathematical models, and data. The system model is a high-level model describing the system and its parameters; we develop this model using the Generic Modeling Environment (GME) platform. Physics-based models, which are usually in the form of equations, are assumed to be in a text format. The data is also assumed to be available in a text format. The proposed methodology involves creating a meta-model for the Bayesian network using GME and a syntax representation for the conditional probability tables/ distributions. The actual Bayesian network is an instance model of the Bayesian network meta-model. We describe algorithms for automated BN construction and UQ analysis, which are implemented programmatically using the GME platform. We finally demonstrate the proposed techniques for quantifying the uncertainty in two example systems.

Gaussian Process Regression (GPR) Representation in Predictive Model Markup Language (PMML)

This paper describes Gaussian process regression (GPR) models presented in predictive model markup language (PMML). PMML is an extensible-markup-language (XML) -based standard language used to represent data-mining and predictive analytic models, as well as pre- and post-processed data. The previous PMML version, PMML 4.2, did not provide capabilities for representing probabilistic (stochastic) machine-learning algorithms that are widely used for constructing predictive models taking the associated uncertainties into consideration. The newly released PMML version 4.3, which includes the GPR model, provides new features: confidence bounds and distribution for the predictive estimations. Both features are needed to establish the foundation for uncertainty quantification analysis. Among various probabilistic machine-learning algorithms, GPR has been widely used for approximating a target function because of its capability of representing complex input and output relationships without predefining a set of basis functions, and predicting a target output with uncertainty quantification. GPR is being employed to various manufacturing data-analytics applications, which necessitates representing this model in a standardized form for easy and rapid employment. In this paper, we present a GPR model and its representation in PMML. Furthermore, we demonstrate a prototype using a real data set in the manufacturing domain.

Model-based Engineering for the Integration of Manufacturing Systems with Advanced Analytics

To employ data analytics effectively and efficiently on manufacturing systems, engineers and data scientists need to collaborate closely to bring their domain knowledge together. In this paper, we introduce a domain-specific modeling approach to integrate a manufacturing system model with advanced analytics, in particular neural networks, to model predictions. Our approach combines a set of meta-models and transformation rules based on the domain knowledge of manufacturing engineers and data scientists. Our approach uses a model of a manufacturing process and its associated data as inputs, and generates a trained neural network model as an output to predict a quantity of interest. This paper presents the domain-specific knowledge that the approach should employ, the formal workflow of the approach, and a milling process use case to illustrate the proposed approach. We also discuss potential extensions of the approach.

A methodology for the semi-automatic generation of analytical models in manufacturing

Abstract Advanced analytics can enable manufacturing engineers to improve product quality and achieve equipment and resource efficiency gains using large amounts of data collected during manufacturing. Manufacturing engineers, however, often lack the expertise to apply advanced analytics, relying instead on frequent consultations with data scientists. Furthermore, collaborations between manufacturing engineers and data scientists have resulted in highly specialized applications that are not relevant to broader use cases. The manufacturing industry can benefit from the techniques applied in these collaborations if they can be generalized for a wide range of manufacturing problems without requiring a strong knowledge about analytical models. This paper first presents a model-based methodology to help manufacturing engineers who have little or no experience in advanced analytics apply machine learning techniques for manufacturing problems. This methodology includes a meta-model repository and model transformations. The meta-models define concepts and rules that are commonly known in the manufacturing industry in order to facilitate the creation of manufacturing models. The model transformations enable the semi-automatic generation of analytical models using a given manufacturing model. Second, a model-based Tool for ADvanced Analytics in Manufacturing (TADAM) is presented to allow manufacturing engineers to apply the methodology. TADAM offers capabilities to generate neural networks for manufacturing process problems. Using TADAM’s graphical user interface, a manufacturing engineer can build a model for a given process to provide: 1) the key performance indicator (KPI) to be predicted, and 2) the variables contributing to this KPI. Once the manufacturing engineer has built the model and provided the associated data, the model transformations available in TADAM can be called to generate a trained neural network. Finally, the benefits of TADAM are demonstrated in a manufacturing use case in which a manufacturing engineer generates a neural network to predict the energy consumption of a milling process.