Data Driven Smart Manufacturing Technologies and Applications

In the past century, the manufacturing industry has undergone significant paradigm shifts, from mass production (Ford assembly line) (1900s) to lean manufacturing (Toyota production system) (1960s), flexible manufacturing (mass customization) (1980s), reconfigurable manufacturing (1990s), collaborative manufacturing (2000s), and smart manufacturing (2010s) [1–3].

Cloud enabled prognosis systems have been increasingly adopted by manufacturing industries. The effectiveness of the cloud systems is, however, crippled by the high latency of data transfer between shop floors and the cloud. To overcome the limitation, this chapter presents an innovative fog enabled prognosis system for machining process optimization. The system functions include: (1) dynamic prognosis - Convolutional Neural Network (CNN) based prognosis is implemented to detect potential faults from customized machining processes. Pre-processing mechanisms of the CNN are designed for partitioning and de-noising monitored signals to strengthen the performance of the system in practical manufacturing situations; (2) an innovative fog enabled prognosis architecture for machining process optimization—it consists of a terminal layer, a fog layer and a cloud layer to minimize data traffic and improve system efficiency. Under the architecture, monitored signals during machining collected on the terminal layer are processed using the trained CNN deployed on the fog layer to efficiently detect abnormal situations. Intensive computing activities like training of the CNN and system re-optimization responding to detected faults are carried out dynamically on the cloud layer to leverage its computation powers. The system was validated in a UK machining company. With the system deployment, the efficiency of energy and production was improved for 29.25% and 16.50% on average. In comparison with a cloud system, this fog system achieved 70.26% reduction in the bandwidth requirement between shop floors and cloud, and 47.02% reduction in data transfer time. This research, sponsored by EU projects, demonstrates that industrial artificial intelligence can facilitate smart manufacturing practices effectively.

The Big Data driven approach has become a new trend for manufacturing optimization. In this chapter, an innovative Big Data enabled Intelligent Immune System (I²S) has been developed to monitor, analyze and optimize machining processes over lifecycles in order to achieve energy efficient manufacturing. There are two major functions in I²S: (1) an Artificial Neural Networks (ANNs)-based algorithm and statistical analysis tools are used to identify the abnormal electricity consumption patterns of manufactured components from monitored Big Data. An intelligent immune mechanism is devised to adapt to the condition changes and process dynamics of machining systems; (2) a re-scheduling algorithm is triggered if abnormal manufacturing conditions are detected thereby achieving multi-objective optimization in terms of energy consumption and manufacturing performance. In this research, Computer Numerical Controlled (CNC) machining processes and industrial case studies have been used for system validation. The novelty of I²S is that Big Data analytics and intelligent immune mechanisms have been integrated systematically to achieve condition monitoring, analysis and energy efficient optimization over manufacturing execution lifecycles. The applicability of the system has been validated by multiple industrial trials in European factories. Around 30% energy saving and over 50% productivity improvement have been achieved by adopting I²S in the factories.

Faults on machines or cutting tooling during machining processes generate negative impacts on productivity, production quality and scrap rate. Effective diagnostics to identify faults throughout the lifecycle of a machining process adaptively is foremost for achieving overall manufacturing sustainability. In recent years, the research of leveraging deep learning algorithms to develop diagnostics approaches has been actively conducted. However, the approaches have not been widely adopted by industries yet due to their inadaptability of addressing the changing working conditions of customized machining processes. Re-collecting a large amount of data and re-training the approaches for new conditions is significantly time-consuming and expensive. To overcome the limitation, this chapter presents a novel deep transfer learning enabled adaptive diagnostics approach. In the approach, firstly, a Long Short-term Memory-Convolutional Neural Network (LSTM-CNN) is designed to perform diagnostics on machining processes. Then, a transfer learning strategy is incorporated into the LSTM-CNN to enhance the adaptability of the approach on different machining conditions via the following steps: (1) The input datasets from different conditions are optimally aligned to facilitate data reuse between the conditions; (2) The weights of the trained LSTM-CNN are regularized using an improved optimization algorithm to minimize the mismatches of feature distributions of the conditions in implementing cross-domain transfer learning. Based on the steps, the LSTM-CNN based diagnosis trained in one condition can be adaptively applied into new conditions efficiently, and thereby the re-training processes of the LSTM-CNN from scratch can be alleviated. Comparative experiment results indicated that the approach achieved 96% in accuracy, which is significantly higher than other approaches without transfer learning mechanisms.

To enhance production quality, productivity and energy consumption, it is paramount to predict the Remaining Useful Life (RUL) of a cutting tool accurately and efficiently. Deep learning algorithm-driven approaches have been actively explored in the research field though there are still potential areas to further enhance the performance of the approaches. In this research, to improve accuracy and expedite computational efficiency for predicting the RUL of cutting tools, a novel systemic methodology is designed to integrate strategies of signal partition and deep learning for effectively processing and analyzing multi-sourced sensor signals throughout the lifecycle of a cutting tool. In more detail, the methodology consists of two subsystems: (i) a Hurst exponent-based method is developed to effectively partition complex and multi-sourced signals along the tool wear evolution; (ii) a hybrid CNN-LSTM algorithm is designed to combine feature extraction, fusion and regression in a systematic means to facilitate the prediction based on segmented signals. The system is validated using a case study with a large set of databases using multiple cutting tools and with multi-sourced signals. Comprehensive comparisons between the proposed methodology and some other main-stream algorithms, such as CNN, LSTM, DNN and PCA, were carried out under the conditions of partitioned and un-partitioned signals. Benchmarks show that, based on the case study in this research, the prediction accuracy of the proposed methodology reached 87.3%, which are significantly better than those of the comparative algorithms.

Heavy-duty CNC machines are important equipment in manufacturing large-scale and high-end products. During the machining processes, a significant amount of heat is generated to bring working temperatures rising, which leads to deformation of machine elements and further machining inaccuracy. In recent years, data-driven approaches for predicting thermal errors have been actively developed to adaptively compensate the errors on the fly to improve machining accuracy. However, it is challenging to adopting the approaches to support heavy-duty CNC machines due to their low efficiency in processing large-volume thermal data. To tackle the issue, this chapter presents a new system for thermal error prediction on heavy-duty CNC machines enabled by a Long Short-Term Memory (LSTM) networks and a fog-cloud architecture. Innovative characteristics of the system include the following aspects: (1) data-based modelling is augmented with physics-based modelling to optimize the number/locations of thermal sensors deployed onto machine elements and minimize excessive data to facilitate computation; (2) a LSTM networks with a data pre-processor is developed for modelling thermal errors more effectively in terms of prediction accuracy and computational efficiency; (3) A fog-cloud architecture is designed to optimize the volume of transferred data and overcome low latency of the system. The system was validated using an industrial heavy-duty CNC machine. Practical case studies show that the system reduced the volume of transmitted data for 52.63% and improved the machining accuracy for 46.53%, in comparison with the processes without using the designed system.

Effective Prognostics and Health Management (PHM) for cutting tools during Computerized Numerical Control (CNC) processes can significantly reduce downtime and decrease losses throughout manufacturing processes. In recent years, deep learning algorithms have demonstrated great potentials for PHM. However, the algorithms are still hindered by the challenge of the limited amount data available in practical manufacturing situations for effective algorithm training. To address this issue, in this research, a transfer learning enabled Convolutional Neural Networks (CNNs) approach is developed to predict the health state of cutting tools. With the integration of a transfer learning strategy, CNNs can effectively perform tool health state prediction based on a modest number of the relevant images of cutting tools. Quantitative benchmarks and analyses on the performance of the developed approach based on six typical CNNs models using several optimization techniques were conducted. The results indicated the suitability of the developed approach, particularly using ResNet-18, for estimating the health state of cutting tools. Therefore, by exploiting the integrated design of CNNs and transfer learning, viable PHM strategies for cutting tools can be established to support practical CNC machining applications.

Adhesive bonded joints are one of important joining technologies in supporting various manufacturing applications. It is important to predict the optimal strength of adhesive bonded joints in order to fit design requirements. Prediction on joint strengths is usually based on experimental tests and Finite Element Analysis (FEA). However, it is a time-consuming and expensive process. To improve computational efficiency and reduce experimental cost, in this chapter, a new approach based on a machine learning algorithm and a FEA method is proposed to predict the failure loads of joints. The innovations of the approach include the following aspects: (1) the FEA model for analyzing the strengths of joints is validated using experimental tests to generate a robust dataset for training a Deep Neural Networks (DNNs) algorithm, which is designed to predict the failure loads of various joint material combinations with high efficiency; (2) based on the trained DNNs algorithm, a Fruit Fly Algorithm (FFO) is proposed to identify the optimal material parameters of Adhesive bonded joints in a given geometry and an joint configuration. The proposed FEA method was successfully validated by conducting experiments with samples of single lap joints. 375 samples were generated by the validated FEA model for DNNs’ training, validation and testing. Case studies showed that the computational time of this approach was saved by 99.54% compared with that of the FEA model, and optimal parameters were identified within 9 iterations based on the proposed FFO optimization algorithm.

Ant Colony Optimization (ACO) is one of the widely used metaheuristic algorithms applicable to various optimization problems. The ACO design has been inspired by the foraging behavior of ant colonies. The performance of the algorithm, however, is not satisfactory since the update stage of pheromones in the algorithm is static and less intelligent, leading to early maturity and slow convergence. To improve the algorithm, in this research, a new ACO algorithm with an innovative adaptive gradient descent strategy (ADACO) is designed, and the algorithm is applied to the Travelling Salesman Problems (TSP) problem for validation. In the chapter, first, the ACO algorithm for TSP is modeled in the framework of stochastic gradient descent (SGD). A new loss function aiming at minimizing the expectation of error ratio and its gradient is defined. Then, an adaptive gradient descent strategy, which can exploit the update history of per-dimensional pheromones to achieve intelligent convergence, is integrated into the ACO algorithm as ADACO. A parallel computation process is also implemented in the algorithm. Finally, ADACO was trialed on various sizes of TSP instances and benchmarked with the Max–Min Ant System (MMAS) algorithm, the Ant Colony System (ACS) algorithm, and the Iterated Local Search (ILS) algorithm. Results show that ADACO outperformed those comparative algorithms in terms of accuracy, stability and adaptability. Furthermore, the results also elucidate that ADACO maintained high-performance computational efficiency owing to its parallel implementation.