Digital Design & Manufacturing
We aim to develop simulation tools with the capability to predict the future behaviour of the assembly system before “physical” installation, commissioning and production. This implies the need to develop a multi-physics defect simulator with capability to model product-related defects and process-induced defects, with the ultimate goal of achieving “right-first-time” capability at (early) design stage.
It has been reported that a leading challenge in delivering cost effective and high quality products is the need to incorporate statistical variation model to tackle product and process variations, so to eliminate/reduce defects during manufacturing stages. Research has demonstrated that around 65–70% of all design changes are related to product variation. It is widely recognized that geometric and dimensional variation are among the most important quality and productivity factors in many assembly processes used not only in automotive (example of large production volume) and aerospace (example of low production volume) but also in appliance, ship-building and other industries. The leading challenges for effective digital design and manufacturing are as follows:
(1)Lack of integration of variation stochastic model to expand current PLM tools to emulate real product/process variations. Integration of stochastic models (i.e., polynomial chaos, Monte Carlo) with nominal CAD/PLM models.
(2)Lack of variation mechanism to model generation and propagation of variations (defects) in multi-stage assembly system. High non-linearity and coupling of design parameters due to stage-to-stage variation propagation mechanism.
(3)Lack of multi-physics modelling capability. Need to expand the capability to detect and isolate defect in multi-physics scenarios (“multi-physics defects”).
(4)Lack of design synthesis strategy. Need to represent the hierarchy of design tasks, used for generating the sequence and importance of parameters/tasks to minimize their interdependencies.
Approach and vision
The key idea is the integration of product and process models along with stochastic mechanisms and first principles approaches (i.e., PDE, ODE) in order to predict the physical behaviour of the multi-stage assembly system in the presence of real manufacturing variations. The ultimate goal is to establish the functional relationship between input process parameters and defect indicators to narrow down defects (i) generated at single stage level (root cause analysis at single-stage level), (ii) and propagated at multi-stage level (root cause analysis at multi-stage level). Items (i) and (ii) are facilitated by (iii) stochastic robust optimisation which include process optimisation with integrated stochastic mechanisms.
We aim to develop closed-loop quality control mechanisms with the ultimate goal of achieving zero-defects manufacturing system. This entail the development of intelligent rapid monitoring and diagnostic methodologies to detect and isolate defects and abnormalities.
Classical approaches for defects diagnostic are mainly concerned with process monitoring to detect the occurrence of any defect, and pinpoint the type of defect and its location within the manufacturing system. Traditional Statistical Control Charts (SPC analysis) have focused on the monitoring of one quality parameters at a time and are not appropriate for analysing process data where variables exhibit collinear behaviour. These limitations are addressed through the application of multivariate statistical process control (M-SPC analysis). M-SPC methods offer the clear advantage to aggregate multiple quality parameters on one individual control chart using T2 and Q statistics. Unfortunately, M-SPC charts also have some disadvantages. The most important one is related to the lack of interpretation of data represented on the control chart. In fact, even though defects can be detected the subsequent isolation is not trivial because of the multiple feasible combinations generating a single fault. The main challenges for rapid diagnostic of defects are enumerated as follows:
(1)Complex non-linear relationship between multiple process parameters and multiple key quality indicators. High non-linearity leads to ill-conditioned systems which may exhibit low level of detectability and isolability of defects.
(2)Heterogeneous & coupled models/parameters. Due to the multi-physics nature of the defects it will be important to tackle coupled and eventually hierarchical dependencies among process parameters and quality defects (“multi-level defects”).
(3)High-dimensional parameter space. When dealing with complex multi-stage assembly systems, the number of actionable and correctable parameters may be very large (up to 1000 control parameters for a single assembly fixture in door assembly systems).
Approach and vision
Each stage of the manufacturing process is understood as a stochastic event with complex interactions between input parts/sub-assemblies, process parameters and variations associated with parts and process parameters. The functional relationship between inputs and outputs is represented as a stochastic multi-variate function, which facilitates defect identification in combination with failure and defects patterns extrapolated on-the-fly with in-process adaptive measurement strategy and rapid diagnostic.
We aim to develop closed-loop quality control mechanisms with the ultimate goal of achieving zero-defects manufacturing system. This entail the development of architecture and models for in-process rapid adjustment and control.
Currently, corrective or preventive actions are mainly data-driven, meaning that they can only work out locally acquired data. Limitations are:
(1)(Partial) open-loop control. Due to lack of (real) in-process measurement systems, process control has to rely on measurements taken at different stages of the manufacturing process and then is unable to deliver automatic closed-loop control strategy.
(2)Accuracy of prediction. Data-driven approaches suffer to be incomplete and inaccurate because they cannot provide a physical reasoning of the phenomena in the assembly system.
(3)Synchronisation to database. Data-driven approaches rely on either static or dynamic database to extrapolate (or interpolate) values to produce results. As a results when comes to extrapolate results outside the database, they suffer to generate low prediction accuracy.
We propose to go from data-driven to model-based approach and to introduce self-optimizing systems into industrial applications that have the ability to correctively and preventively act on the process based on model-based reasoning information. This underlies the following challenges:
(1)Gathering of heterogeneous data structures. Both spatial and temporal data needed to diagnose defects. Spatial data allow to isolate defects within the single-stage; temporal data are necessary for defect localisation across assembly stages.
(2)Computational time. The real-time fault diagnosis demands efficient algorithms which generate actionable results in due time, depending on the production rate and the assembly operation.
(3)Feed-back to design models based on in-process data. Preventing defects and optimizing process require the feed-back of in-process data into product/process design which currently is not fully exploited.
Approach and vision
Our vision is to correct and prevent only those defects which trigger unwanted deterioration of product quality. This is facilitated by novel algorithms with the capability to model: (i) multi-physics defect simulation; (ii) intelligent root cause analysis evaluation; and (iii) intelligent knowledge repository based on advanced machine learning models, such as neural network, rough set, fuzzy logic etc.
Industrial Systems - Application for Healthcare System
To develop systematic methodologies and simulation tools through the integration of care processes, resources and complex service system information to reduce and prevent unnecessary variation in care along the patient journey for quality improvement.
Healthcare organizations represent fast-paced environments that are often characterised by high levels of uncertainties. Depending upon patient conditions, multiple roles, therapies, resources and quality indicators interact with the care journey to ensure the delivery of appropriate care. Often the roles and resources are not clearly defined and may be subject to frequent changes. Also, the progression of the care journey generates data that are heterogeneous in nature and reside in multiple locales that are not always connected. Frequently these data are ill-structured, uncertain, and/or missing. Taken together, these factors trigger unnecessary variation in care that must be reduced and where possible prevented. Some of the key challenges towards reducing such variation may be seen as:
(1) Lack of approaches to accurately model patient journey through the service system
(2) Lack of approaches to model the propagation of unnecessary variation in patient journey
(3) Lack of methods to model the complex relationship between care-related processes with quality indicators
Approach and vision
Patient care is essentially process driven. Plus, the healthcare system is both resource intensive and measured through numerous quality indicators. Research activities aim to develop systematic methodologies by integrating process related measurements with key performance indicator data involving care quality requirements and service performance to understand, reduce and prevent unnecessary variations in care in the patient journey through the service system.