Andrés Boza¹, Pedro Gomez-Gasquet¹ , David Perez¹ and Faustino Alarcón¹

¹ Çentro de Investigación Gestión e Ingeniería de la Producción (CIGIP), Universitat Politècnica de València, Spain.

aboza@cigip.es, pgomez@cigip.es, dapepe@cigip.es, faualva@cigip.es

Keywords: Production Planning, Mathematical model, Data Model.

1. Introduction

Model-Driven Decision Support System help decision-makers to make planning decisions useful for a period of time. Mathematical models for production planning facilitate the decision-making for a better organization of the production according to certain criteria and business restrictions. However, the digital transformation of organizations together with the promotion of new technologies in the field of Industry 4.0 means that these models must be revised to be adapted to this new industrial reality.

Deciding which mathematical model best fits the reality (and business need) is not an easy task. On the one hand, deciding in the conceptualization which elements of the organization are relevant and should participate in the model is not easy since the scope of the problem must be limited without leaving out any aspect of interest to it. On the other hand, establishing the relationships between the elements of the model as well as establishing the indicator or indicators that allow the decisions obtained to be compared is not easy either. Thus, we can find controllable factors (which can be set within a range) and uncontrollable factors.  Also, production planning can be affected by different unexpected events or incidences, for example, a broken machine or a huge order. If the detection of the event is slow, the troubles will be bigger. In this sense, new technologies like Internet of Things can help in this purpose. A quicker identification of relevant events is necessary to make a quicker analysis of their consequences. All this set of possibilities in the design of the models make necessary tools for their analysis.

2. Proposal for a model experimentation environment

The proposal focusses on the analysis of the decision models. That is, the search of better mathematical models to be used later in production contexts. Experimenting with the models to achieve higher quality can produce results: i) Reliable (about how close repeated measurements are to each other); ii) Accuracy (how close the final result is to the correct or accepted value). Both are affected by the time limit that we let the algorithm or resolution engine work. Also, it can be affected by the different data instances used.

The proposed design includes three subsystems: Data Modeling, Decision Modeling and Model Analysis and Investigation:

Data Modelling refers to the ‘structured’ internal representation and external presentation of recorded facts. Broadly speaking, it provides the decision-maker with information about their decision problem. Decision Modelling is the development of a model, or a range of models that captures the structure as well as the decisions in respect of a given problem. These models are used to evaluate possible decisions (actions) in a given problem domain, and the probable outcomes of these actions. And, Model Analysis and Investigation refers to the instantiation of the model with data, and the evaluation of the model parameters as well as the results in order to gain confidence and insight into the model. That is, the design seeks a tool that facilitates design of experiments, making strategic and deliberate changes to produce useful information for the improvement in the models. A first approach in the search for quality models includes the following steps:

1. Model Analysis: Syntax error checking can detect ‘early’ anomalies in the model formulation.
2. Model and data validation. The analyst checks whether or not the model makes sense with the model validation.
3. Solution analysis and investigation. After model diagnosis, the analyst may carry out ‘what-if’ analyses (or scenarios analysis), where the analyst changes the input values, using different model data instances. Also, the diagnosis can guide the introduction of change in the models.

The design of the proposal at a more detailed level includes the Data Modeling subsystems which structures the data necessary for the decision models and is connected to the information sources from which to extract the data from different scenarios through ETL processes. The Decision Modeling for the inclusion and storage of the decision models according to the Data modeling, and the Model Analysis and Investigation subsystem for the analysis of the resolution of the models with the data instances used in the experimentation environment.

3. Conclusions

The need to adapt the production planning models to the new Industry 4.0 environments justifies the proposal for the improvement and validation of new decision models. The proposed design facilitates the experimentation of models to later be exploited in business environments. The design allows the model design-er together with the decision-maker to validate the usefulness of the models and adjust them to their business reality. The main advantage lies in the existence of a reusable model experimentation environment for different experiments. Another advantages are: a) the versatility in the experimentations to use different data models and decision models, b) the separation of data instantiation from the data model design and the decision model, and c) the control of the results of the different scenarios proposed in each experiment.

Acknowledgements. This research has been funded by the project entitled NIOTOME (Ref. RTI2018-102020-B-I00) (MCI/AEI/FEDER, UE)