## Using Bayesian Networks in an Industrial Setting: Making Printing Systems Adaptive

Citations: | 1 - 1 self |

### BibTeX

@MISC{Hommersom_usingbayesian,

author = {Arjen Hommersom and Peter J. F. Lucas},

title = {Using Bayesian Networks in an Industrial Setting: Making Printing Systems Adaptive},

year = {}

}

### OpenURL

### Abstract

Abstract. Control engineering is a field of major industrial importance as it offers principles for engineering controllable physical devices, such as cell phones, television sets, and printing systems. Control engineering techniques assume that a physical system’s dynamic behaviour can be completely described by means of a set of equations. However, as modern systems are often of high complexity, drafting such equations has become more and more difficult. Moreover, to dynamically adapt the system’s behaviour to a changing environment, observations obtained from sensors at runtime need to be taken into account. However, such observations give an incomplete picture of the system’s behaviour; when combined with the incompletely understood complexity of the device, control engineering solutions increasingly fall short. Probabilistic reasoning would allow one to deal with these sources of incompleteness, yet in the area of control engineering such AI solutions are rare. When using a Bayesian network in this context the required model can be learnt, and tuned, from data, uncertainty can be handled, and the model can be subsequently used for stochastic control of the system’s behaviour. In this paper we discuss industrial research in which Bayesian networks were successfully used to control complex printing systems. 1

### Citations

2535 |
A new approach to linear filtering and prediction problems
- Kalman
- 1960
(Show Context)
Citation Context ... model. Typical applications are filtering – i.e., inferring the current hidden state given the observations in the past – and smoothing where past states are inferred. For example, the Kalman filter =-=[7]-=- is well-known in stochastic control theory (see e.g., [1]) and is a special case of a dynamic Bayesian networks, where the model is the linear Gaussian variant of a hidden Markov model, i.e., it desc... |

1355 |
Local computations with probabilities on graphical structures and their application to expert systems
- Lauritzen, Spiegelhalter
- 1988
(Show Context)
Citation Context ...commonly used type of hybrid Bayesian network is the conditional linear Gaussian model [3, 8]. Efficient exact and approximate algorithms have been developed to infer probabilities from such networks =-=[10, 2, 12]-=-. Also important in the context of embedded systems is that real-time probabilistic inference can be done using anytime algorithms [6]. A Bayesian network can be constructed with the help of one or mo... |

1028 | Monte Carlo Statistical Methods
- Robert, Casella
- 1999
(Show Context)
Citation Context ...commonly used type of hybrid Bayesian network is the conditional linear Gaussian model [3, 8]. Efficient exact and approximate algorithms have been developed to infer probabilities from such networks =-=[10, 2, 12]-=-. Also important in the context of embedded systems is that real-time probabilistic inference can be done using anytime algorithms [6]. A Bayesian network can be constructed with the help of one or mo... |

515 | Sequential Monte Carlo methods for dynamic systems
- Liu
- 1998
(Show Context)
Citation Context ...tended Kalman filter or the unscented Kalman filter (see e.g., [15]) are approximate inference algorithms for non-linear Gaussian models by linearisation of the model. More recently, particle filters =-=[13]-=-, have been proposed as an alternative, which relies on sampling to approximate the posterior distribution.The difference with these filtering approaches is that for Bayesian networks there is an und... |

339 |
Stochastic Models, Estimation, and Control
- Maybeck
- 1979
(Show Context)
Citation Context ...v model, i.e., it describes a Markov process with noise parameters on the input and output variables. Non-linear variants, such as the extended Kalman filter or the unscented Kalman filter (see e.g., =-=[15]-=-) are approximate inference algorithms for non-linear Gaussian models by linearisation of the model. More recently, particle filters [13], have been proposed as an alternative, which relies on samplin... |

232 |
The EM algorithm for graphical association models with missing data
- Lauritzen
- 1995
(Show Context)
Citation Context ...eter learning. This is typically done by computing the maximum likelihood estimates of the parameters, i.e., the conditional probability distributions, associated to the networks structure given data =-=[9]-=-. Temporal Bayesian networks are Bayesian network where the vertices of the graph are indexed with (discrete) time. All vertices with the same time index form a so-called time slice. Each time slice c... |

150 |
Introduction to Stochastic Control Theory
- Astrom
- 1970
(Show Context)
Citation Context ...ng the current hidden state given the observations in the past – and smoothing where past states are inferred. For example, the Kalman filter [7] is well-known in stochastic control theory (see e.g., =-=[1]-=-) and is a special case of a dynamic Bayesian networks, where the model is the linear Gaussian variant of a hidden Markov model, i.e., it describes a Markov process with noise parameters on the input ... |

148 | Propagation of probabilities, means, and variances in mixed graphical association models
- Lauritzen
- 1992
(Show Context)
Citation Context ...rks, however, containing both discrete and continuous conditional probability distributions are also possible. A commonly used type of hybrid Bayesian network is the conditional linear Gaussian model =-=[3, 8]-=-. Efficient exact and approximate algorithms have been developed to infer probabilities from such networks [10, 2, 12]. Also important in the context of embedded systems is that real-time probabilisti... |

65 | 2001), “Inference in hybrid networks: Theoretical limits and practical algorithms
- Lerner, Parr
(Show Context)
Citation Context ...commonly used type of hybrid Bayesian network is the conditional linear Gaussian model [3, 8]. Efficient exact and approximate algorithms have been developed to infer probabilities from such networks =-=[10, 2, 12]-=-. Also important in the context of embedded systems is that real-time probabilistic inference can be done using anytime algorithms [6]. A Bayesian network can be constructed with the help of one or mo... |

46 |
Software-engineering research revisited
- Potts
- 1928
(Show Context)
Citation Context ...eadable Bayesian networks. 6 CONCLUSIONS In embedded software, there is an increasing trend to apply and verify new software methods in an industrial context, i.e., the industry-aslaboratory paradigm =-=[18]-=-. This means that concrete cases are studied in their industrial context to promote the applicability and scalability of solution strategies under the relevant practical constraints. Much of the curre... |

36 | A survey of algorithms for real-time Bayesian network inference
- Guo, Hsu
(Show Context)
Citation Context ... been developed to infer probabilities from such networks [10, 2, 12]. Also important in the context of embedded systems is that real-time probabilistic inference can be done using anytime algorithms =-=[6]-=-. A Bayesian network can be constructed with the help of one or more domain experts. However, building Bayesian networks using expert knowledge, although by now known to be feasible for some domains, ... |

32 | Dynamic Bayesian Networks
- MURPHY
- 2002
(Show Context)
Citation Context ...cture and parameters of the static Bayesian network are the same at every time slice (with the exception of the first), one speaks of a dynamic Bayesian network, as such networks can be unrolled (cf. =-=[16]-=- for an overview). 2.2 Bayesian-network modelling of a dynamic system Common assumptions in the modelling of dynamic physical systems are that the system is Markovian and stationary (e.g., [11]), i.e.... |

25 |
Computerbased decision-support in the management of primary gastric non-Hodgkin lymphoma. Meth Inform Med
- PJF, Boot, et al.
- 1998
(Show Context)
Citation Context ...hat the structure of the graphical part of a Bayesian network facilitates the assessment of probabilities, even to the extent that reliable probabilistic information can be obtained from experts (see =-=[14]-=-). One other advantage compared to black-box models is that the modelled probability distribution can be exploited for decision making using decision theory. This is particularly important if one want... |

22 | Monitoring a complex physical system using a hybrid dynamic bayes net
- Lerner, Moses, et al.
- 2002
(Show Context)
Citation Context ...d (cf. [16] for an overview). 2.2 Bayesian-network modelling of a dynamic system Common assumptions in the modelling of dynamic physical systems are that the system is Markovian and stationary (e.g., =-=[11]-=-), i.e., the system state at time t + 1 is only dependent on the system state at time t, and the probabilistic dependencies are the same at each time t. Stationarity is an assumption too strong for th... |

10 |
Adaptive Approximation Based Control: Unifying Neural, Fuzzy and Traditional Adaptive Approximation Approaches
- Farrell, Polycarpou
- 2006
(Show Context)
Citation Context ...f artificial intelligence, such as rule-based systems, fuzzy logic, neural networks, evolutionary algorithms, etc. have been used in order to determine optimal values for control parameters (see e.g. =-=[5]-=-). The work presented in this paper extends these approaches using human-readable Bayesian networks. 6 CONCLUSIONS In embedded software, there is an increasing trend to apply and verify new software m... |

8 |
Probabilistic reasoning in inteligent systems: Networks of plausible infernece
- Pearl
- 1988
(Show Context)
Citation Context ... most appropriate. 1 Radboud University Nijmegen, Institute for Computing and Information Sciences, The Netherlands, email: {arjenh,peterl}@cs.ru.nl In this paper, we propose to use Bayesian networks =-=[17]-=- to deal with the control of such complex systems. The formalism possesses the unique quality of being both an AI-like and statistical knowledgerepresentation formalism. Nowadays, Bayesian networks ta... |

3 |
Modeling and control of static and dynamic systems with Bayesian networks
- Deventer
- 2004
(Show Context)
Citation Context ...ity and energy consumption. With respect to decision making, adaptive controllers using explicit Bayesian networks have not been extensively investigated. The most closely related work is by Deventer =-=[4]-=-, who investigated the use of dynamic Bayesian networks for controlling linear and nonlinear systems. The premise of this work is that the parameters of a Bayesian network can be estimated from a dete... |

2 |
Uffe Kjærulff, ‘Printer troubleshooting using Bayesian networks
- Skaanning, Jensen
- 2000
(Show Context)
Citation Context ... they could also be used or re-used for diagnostic purposes, where it is typically required that a diagnosis can be represented in a human-understandable way so that proper action can be taken (e.g., =-=[19]-=- in the printing domain). Furthermore, it is well-known that the structure of the graphical part of a Bayesian network facilitates the assessment of probabilities, even to the extent that reliable pro... |