Data-Driven Process Improvement

Manufacturing companies are aware of the importance of data collection and data analytics to develop their process knowledge and to remain competitive. Nevertheless, data analytics has recently been gaining more attention, since it plays a significant role in streamlining manufacturing operations to be faster and more efficient, especially in the domain of smart manufacturing. In particular, the benefits of data processing/analytics are not limited to visualising the status of the manufacturing processes but also empower artificial intelligence (AI) algorithms to produce smart solutions and concurrent corrective actions (see Section 3.7.2). Currently, manufacturing data is utilised to model AM processes and to facilitate the optimisation of the process parameters. Data can be gathered via sensors in real time (see Section 3.7.1 and Chapter 13) or from off-line experimental measurements (see Section 3.6.1, 3.6.2 and Chapters 10,11 and 12).

A data-driven process improvement is structured based on five steps (collection of data, sharing, analysis, optimisation and feedback), as shown in Figure 3.9.

In a given AM process, data can be collected manually via an operator or automatically via smart sensors. AM data is highly heterogeneous as it is collected from different sources, such as sensors, cameras, databases and measurements of experimental tests. The format of the collected data has to be digitised and normalised to render it compatible with the modelling and optimisation algorithms. In addition, the digital format makes the data shareable between all smart assets in the manufacturing system (server machine, control units, monitoring tools, inspection operator and process documentation). Data collection and data sharing are followed by a further data analysis step to extract and visualise the mathematical model/objective function of the process. The data-analysis algorithms need to conduct an additional step for data cleaning/delet- ing in order to prevent data redundancy and data inconsistency (DeCastro-Garda et al. 2018, Rahm and Do 2000). Machine learning algorithms can be utilised to simplify the analysis via parallel processing of the data without necessitating a high level of computing power. Machine learning algorithms also reduce the bandwidth for sharing data and handle issues such as missing information and communication failures (Dean and


Data-driven process improvement steps. (Modified from Buer et al. 2018.)

Ghemawat 2004). The analysis step produces a deterministic mathematical function that models the relationship between all processed data. This model/mathematical function is used to predict/describe the behaviour of the AM process under various conditions (Banerjee et al. 2014, Mertens et al. 2014, Tapia et al. 2016). Following the mathematical modelling, the optimisation step identifies the optimal values of the process parameters for the best possible performance of the AM process. In particular, the optimisation step adjusts the parameters of the AM model in order to maximise the precision/qual- ity of the product or minimise manufacturing errors/costs, depending on the objective function. Finally, the feedback step feeds the machine/operator the corrective actions/ parameters in order to continuously improve the process.

< Prev   CONTENTS   Source   Next >