1,720,981 research outputs found
Scalable Approximate Optimization of Objective Functions Represented by Random Forests
Get PDFThe problem of global optimization of an objective function represented by a Random Forest (RF) is considered. A method to obtain an approximate solution at low computational complexity is proposed, resorting to the inherent structure of an RF, which is a non-parametric model that partitions the feature space in convex polytopes according to the training data. The approach selects the optimal solution inside the polytopes corresponding to the best data points. It is shown that the proposed approximate method is significantly more efficient, thus applicable at large scale, than extensive global search algorithms, such as gridding and Mixed Integer Linear Programming (MILP), which in turn provide exact solutions. The efficiency and sub-optimality of the approach are evaluated on RFs trained on a dataset generated by sampling a bivariate, discontinuous and non-convex benchmark function from the literature
Model-based adaptive process control for surface finish improvement in traverse grinding
Get PDFThe paper presents a process controller aimed at improving the surface quality generated by traverse grinding, avoiding the surface defects caused by vibrations onset. The innovation provided by the proposed controller consists in suppressing vibration occurrence by means of a model-based and self-learning approach: a monitoring layer classifies occurring problems and a control logic exploits these indications to select the proper mitigation actions. Since wheel-regenerative chatter represents one of the most important problems during traverse grinding in terms of achievable productivity and finishing quality, the main control variable is the wheel velocity. This variable is tuned exploiting an adaptive Speed tuning Map computed by the controller using a heuristic approach and learning methodology. The control can manage also the other sources of vibration by means of proper identification and mitigation strategies. Experimental tests are carried out on a roll grinder to validate the control system. Good performances are achieved after some training tests to allow controller learning
On Development of an Optimal Control System for Real-time Process Optimization on Milling Machine Tools
No full textDeveloping an intelligent machine tool means to augment its level of automation. This augmentation, in turn, requires a machine
controller able to perform actions and to implement attributes that are currently demanded to, and hold by, human operators. The
present paper describes how this issue is being faced by a large Italian national research project, funded under the Industria 2015
initiative. Considering the case of milling machines, human operators are currently in charge of supervising the cutting process by
acting on spindle speed and feed override controls in order to compensate for undesired process conditions (e.g. excessive vibration
or power absorption) caused by a wrong choice of process parameters during the design of the part-program, by tool wear, by
unexpected work material properties, or by machine tool dynamics. The first part of the paper proposes the architecture for an
augmented-automation machine tool. Rather than revolutionizing the well-established architecture of a conventional machine tool,
the concept is based on an additional controller that implements a supervision and optimization loop. This additional controller gets
process state information from the CNC and from dedicated measurement systems, and closes a feedback action on the CNC as a
human operator would do: by acting on feed and spindle speed overrides. The second part of the paper illustrates how the additional
controller works: following the optimal control theory, it is based on a dynamic process model, a set of state variables (i.e.
measurements), and a set of controls. Exploiting a simplified process model and efficient optimization algorithms, it performs a
real-time optimization of the controls (i.e. the overrides named above) on the basis of a weighted multi-objective target function and
a set of measurements taken from the cutting process (power, forces, accelerations). In particular, the target function takes into
account the following objectives: cutting time, work-piece surface finish, tool wear rate and vibration mitigation in general. The
third part of the paper details the strategies concerned with tool vibrations prediction, monitoring and mitigation, which are
integrated into the optimization loop. A vibrations prediction module based on a simplified cutting process model allows the
estimation of the vibration level and/or chatter occurrence during a pre-processing phase: thus, through the computation of the
Stability Lobes Diagram along the tool path, the more stable spindle speeds can be identified. The pre-processing phase is
complemented with an in-process chatter monitoring algorithm based on a recursive dynamic model identification: detecting real-
time self-excited vibrations onset, and distinguishing them from forced vibrations, this module allows the controller to properly
update the vibration estimation
Prediction of power consumption from real process data of an industrial wood chip refining plant
Get PDFThe energy issue leads to improve the efficiency of production processes, which plays a fundamental role in minimizing their environmental impact and energy consumption. The pulp
and paper industry is an excellent representation of an energy-intensive manufacturing process that urgently needs to become more efficient. In particular, the primary focus should be placed on the refining phase, which constitutes most of the energy usage. This characteristic is shared by all process that includes wood fiber extraction, such as the engineered wood panel industry. Within this frame, our work deals with the identification of a
wood chips refining process operating in closed loop, committed to the production of Medium Density Fiberboard (MDF), in order to provide a long-term prediction of energy consumption. We perform the identification via multi-batch Simulation Error Minimization (SEM) from a real process data set related to a large-scale production plant. In conclusion, we will propose the application of the refiner motor current estimation and of the refiner disc gap drift to model the wear effect that characterizes the process
Force-field instability in surface grinding
No full textIn this paper, a particular kind of non-regenerative instability in surface grinding is studied. Clear evidences have been collected suggesting that vibrations can occur suddenly even during the first grinding pass, just after wheel dressing. These circumstances exclude workpiece and wheel surface regeneration as instability origin, whereas both surfaces have to be considered initially smooth. On these bases, the stability of the dynamic system constituted by an oscillating ideal wheel (namely without waviness on the surface) immerged in a positional and velocity-dependent process force field has been studied, demonstrating that, under particular conditions, the force field generates an unstable behaviour. The instability occurrence is strictly related to the oscillation direction of the wheel centre, according to the mode shape associated to the dominant resonance, with respect to the direction of the grinding force (identified by the ratio between its tangential and normal components). The analysis leads to the identification of a simple necessary condition for instability occurrence. The analytical results are confirmed by time-domain grinding simulations and compared with experimental evidences
An integrated approach for joint process planning and machine tool dynamic behavioral assessment
No full textThe current work introduces a novel approach for enhanced process planning where the machine tool kinematic and dynamic behavior is introduced and modeled. The proposed approach is structured in three main steps starting from the workpiece analysis, the dynamic cutting simulation till the fixturing system selection and the setup planning. The methodology introduced in the paper has been validated with reference to an industrial case study and results have been described
Energy driven process planning and machine tool dynamic behavior assessment
No full textThe current work outlines an approach to close the loop between process planning and machine tool dynamic modeling by addressing the problem of energy efficiency across the process design and realization chains, from the process settings and pallet configuration to the machine tool design and usage phases. The proposed closed loop approach consists of an off-line and on-line component enabling the process and equipment dynamic and energy assessment over time. The benefits of the approach have been evaluated against an industrial case study related to the automotive industry
Roundness prediction in centreless grinding using physics-enhanced machine learning techniques
Get PDFThis work proposes a model for suggesting optimal process configuration in plunge centreless grinding operations. Seven different approaches were implemented and compared: first principles model, neural network model with one hidden layer, support vector regression model with polynomial kernel function, Gaussian process regression model and hybrid versions of those three models. The first approach is based on an enhancement of the well-known numerical process simulation of geometrical instability. The model takes into account raw workpiece profile and possible wheel-workpiece loss of contact, which introduces an inherent limitation on the resulting profile waviness. Physical models, because of epistemic errors due to neglected or oversimplified functional relationships, can be too approximated for being considered in industrial applications. Moreover, in deterministic models, uncertainties affecting the various parameters are not explicitly considered. Complexity in centreless grinding models arises from phenomena like contact length dependency on local compliance, contact force and grinding wheel roughness, unpredicted material properties of the grinding wheel and workpiece, precision of the manual setup done by the operator, wheel wear and nature of wheel wear. In order to improve the overall model prediction accuracy and allow automated continuous learning, several machine learning techniques have been investigated: a Bayesian regularized neural network, an SVR model and a GPR model. To exploit the a priori knowledge embedded in physical models, hybrid models are proposed, where neural network, SVR and GPR models are fed by the nominal process parameters enriched with the roundness predicted by the first principle model. Those hybrid models result in an improved prediction capability
Frequency domain identification of grinding stiffness and damping
Get PDFAs equivalent stiffness and damping of the grinding process dominate cutting stability, their identification is essential to predict and avoid detrimental chatter occurrence. The identification of these process constants is not easy in large cylindrical grinding machines, e.g. roll grinders, since there are no practical ways to measure cutting force normal component. This paper presents a novel frequency domain approach for identifying these process parameters, exploiting in-process system response, measured via impact testing. This method adopts a sub-structuring approach to couple the wheel-workpiece relative dynamic compliance with a two-dimensional grinding force model that entails both normal and tangential directions. The grinding specific energy and normal force ratio, that determine grinding stiffness and damping, are identified by fitting the closed loop FRF (Frequency Response Function) measured during specific plunge-grinding tests. The fitting quality supports the predictive capability of the model. Eventually, the soundness of the proposed identification procedure is further assessed by comparing the grinding specific energy identified through standard cutting power measurements
Parametric and Non-Parametric Identification of Micromilling Dynamics
No full textMonitoring and control of micromilling process represents a challenging task in metal cutting. The lack of static and dynamic stiffness due to the small tool size represents one of the most important issues that has to be tackled in order to provide satisfactory control of the process. In this paper different methods for identifying micro-end milling dynamics for process monitoring are proposed. On one side a parametric approach based on the identification of machining system ODPs (Operation Dynamic Parameters) has been designed. On the other hand a non-parametric approach, based on the calculation of system Lyapunov exponents has been also tested to verify whether generalized methods for assessing system dynamics are also applicable in micromachining where the process nonlinearity’s can become relevant, thus limiting the effectiveness of other monitoring methods
- …
