Online Change Detection Research Articles

Online and Non-Parametric Drift Detection Methods Based on Hoeffding’s Bounds

Incremental and online learning algorithms are more relevant in the data mining context because of the increasing necessity to process data streams. In this context, the target function may change over time, an inherent problem of online learning (known as concept drift). In order to handle concept drift regardless of the learning model, we propose new methods to monitor the performance metrics measured during the learning process, to trigger drift signals when a significant variation has been detected. To monitor this performance, we apply some probability inequalities that assume only independent, univariate and bounded random variables to obtain theoretical guarantees for the detection of such distributional changes. Some common restrictions for the online change detection as well as relevant types of change (abrupt and gradual) are considered. Two main approaches are proposed, the first one involves moving averages and is more suitable to detect abrupt changes. The second one follows a widespread intuitive idea to deal with gradual changes using weighted moving averages. The simplicity of the proposed methods, together with the computational efficiency make them very advantageous. We use a Naive Bayes classifier and a Perceptron to evaluate the performance of the methods over synthetic and real data.

Monitoring and condition-based maintenance with abrupt change in a system’s deterioration rate

In this paper, the maintenance of a system which undergoes a change in its deterioration rate is considered. The time of change of regime and the parameters after the change are unknown. To deal with unknown change time and unknown parameters, a detection procedure based on a suitable online change detection algorithm is used. The paper proposes a maintenance decision rule versus detection policy in order to minimise the long-run average maintenance cost, and the performances of this policy are studied through numerical implementations.

Change detection in urban areas by object-based analysis and on-the-fly comparison of multi-view ALS data

The use of helicopters as a sensor platform offers flexible fields of application due to adaptable flying speed at low flight levels. Modern helicopters are equipped with radar altimeters, inertial navigation systems (INS), forward-looking cameras and even laser scanners for automatic obstacle avoidance. If the 3D geometry of the terrain is already available, the analysis of airborne laser scanner (ALS) measurements may also be used for terrain-referenced navigation and change detection. In this paper, we present a framework for on-the-fly comparison of current ALS data to given reference data of an urban area. In contrast to classical difference methods, our approach extends the concept of occupancy grids known from robot mapping. However, it does not blur the measured information onto the grid cells. The proposed change detection method applies the Dempster–Shafer theory to identify conflicting evidence along the laser pulse propagation path. Additional attributes are considered to decide whether detected changes are of man-made origin or occurring due to seasonal effects. The concept of online change detection has been successfully validated in offline experiments with recorded ALS data streams. Results are shown for an urban test site at which multi-view ALS data were acquired at an interval of 1year.

Online change detection of Markov chains with unknown post-change transition probabilities

ABSTRACTIn this paper, we investigate the performance of cumulative sum (CUSUM) stopping rules for the online detection of unknown change point in a time homogeneous Markov chain. Under the condition that the post-change transition probabilities are unknown, we proposed two CUSUM type schemes for the detection. The first scheme is based on the maximum likelihood estimates of the post-change transition probabilities. This scheme is limited by its computation burden, which is mitigated by another scheme based on the reference transition probabilities selected from a prior known region. We give the bounds of the mean delay time and the mean time between false alarms to illustrate the effectiveness of the proposed schemes. The results of the simulation also demonstrate the feasibility of the proposed schemes.

On-line change detection and condition-based maintenance for systems with unknown deterioration parameters

On-line change detection and condition-based maintenance for systems with unknown deterioration parameters

Detection of traffic changes in large-scale backbone networks: The case of the Spanish academic network

Network management systems produce a huge amount of data in large-scale networks. For example, the Spanish academic network features hundreds of access and backbone links, each of which produces a link utilization time series. For the purpose of detecting relevant changes in traffic load a visual inspection of all such time series is required. As a result, the operational expenditure increases. In this paper, we present an on-line change detection algorithm to identify the relevant change points in link utilization, which are presented to the network manager through a graphical user interface. Consequently, the network manager only inspects those links that show a stationary and statistically significant change in the link load.

Condition-based maintenance for a system subject to a non-homogeneous wear process with a wear rate transition

The aim of this paper is to propose an adaptive maintenance model for a gradually deteriorating system. The system considered initially deteriorates with a nominal deterioration rate and at an unknown time the system's deterioration rate changes and the new deterioration rate is a time-dependent function. To deal with the transition of mode of deterioration in the framework of the maintenance decision rule an adequate online change detection algorithm is used. The maintenance decision rule is chosen in order to minimise the total maintenance cost including the cost of unavailability. The main result of this paper is to point out the interest of using a detection algorithm and hence the appreciation of a decision rule which takes into account transitions in the deterioration rate.

Information theoretic novelty detection

We present a novel approach to online change detection problems when the training sample size is small. The proposed approach is based on estimating the expected information content of a new data point and allows an accurate control of the false positive rate even for small data sets. In the case of the Gaussian distribution, our approach is analytically tractable and closely related to classical statistical tests. We then propose an approximation scheme to extend our approach to the case of the mixture of Gaussians. We evaluate extensively our approach on synthetic data and on three real benchmark data sets. The experimental validation shows that our method maintains a good overall accuracy, but significantly improves the control over the false positive rate.

A Change Detection Algorithm based on Recursive Subspace Identification and its Application to a Cart System

This paper presents an on-line change detection method that uses a recursive subspace model identification algorithm as a residual generator. In our method, a change in the variance of the residual is monitored by a statistical change detector based on the log-likelihood ratio test. An experiment using a real-life cart system as well as a numerical example demonstrates that the proposed method can detect changes in the dynamics of a system, without being disturbed by changes in the dynamics of an input signal which are not our concern.

Maintenance decision rule with embedded online Bayesian change detection for gradually non-stationary deteriorating systems

The aim of a condition-based maintenance policy is to manage the available online information about a component or a (sub)system, usually its degradation level, in order to improve the maintenance decision-making. This paper tackles the problem of maintenance decision rules for stochastically deteriorating systems that are subject to changes of their degradation rate during a life cycle. A well-suited control-limit maintenance decision rule is considered with an embedded online change detection algorithm. The maintenance decision and change detection parameters are optimized with respect to the same global maintenance cost and according to the available information about the degradation process. The obtained policy is compared with more classical control limit condition-based maintenance policies without online change detection. The use of the embedded online detection algorithm shows its efficiency especially for systems subject to significantly different degradation rates.

On the use of on-line detection for maintenance of gradually deteriorating systems

This paper deals with condition-based maintenance and non-stationary degradation process due to sudden changes. This is an attempt to propose an adaptive maintenance policy based on the on-line change detection procedure which can help to detect switches from a nominal mode to an accelerated mode in a non-informative context about the change mode time.

Online change detection and condition-based maintenance

The aim of this paper is to use the online change detection/isolation methods in the framework of the condition-based maintenance. The purpose is to propose an adequate condition-based maintenance policy to a gradually deteriorating system with change of mode using on-line detection algorithms. The parameters defining the deterioration mode after the change can be unknown. The main purpose is to optimize a global cost criterion.

Distributed Bayesian fault diagnosis of jump Markov systems in wireless sensor networks

A Bayesian distributed online change detection algorithm is proposed for monitoring a dynamical system by a wireless sensor network. The proposed solution relies on modelling the system dynamics by a jump Markov system with a finite set of states, including the abrupt change behaviour. For each discrete state, an observed system is assumed to evolve according to a state-space model. The collaborative strategy ensures the efficiency and the robustness of the data processing, while limiting the required communications bandwith. An efficient Rao-Blackwellised Collaborative Particle Filter (RB-CPF) is proposed to estimate the a posteriori probability of the discrete states of the observed systems. The Rao-Blackwellisation procedure combines a Sequential Monte-Carlo (SMC) filter with a bank of distributed Kalman filters. In order to prolong the sensor network lifetime, only few active (leader) nodes are selected according to a spatio-temporal selection protocol. This protocol is based on a trade-off between error propagation, communications constraints and information content complementarity of distributed data. Only sufficient statistics are communicated between leader nodes and their collaborators.

The standard Poisson disorder problem revisited

A change in the arrival rate of a Poisson process sometimes necessitates immediate action. If the change time is unobservable, then the design of online change detection procedures becomes important and is known as the Poisson disorder problem. Formulated and partially solved by Davis [Banach Center Publ., 1 (1976) 65–72], the standard Poisson problem addresses the tradeoff between false alarms and detection delay costs in the most useful way for applications. In this paper we solve the standard problem completely and describe efficient numerical methods to calculate the policy parameters.

A BAYESIAN ON-LINE CHANGE DETECTION ALGORITHM WITH PROCESS MONITORING APPLICATIONS

This article presents a Bayesian on-line change detection algorithm for the cases when there are multiple jumps and when there is a trend in the process output parameter. A decision-theory-based method has been formulated to determine the optimum inspec..

Robust control of robot manipulators with fast load adaptation

Many adaptive robot controllers have been proposed in the literature to solve manipulator trajectory tracking problems for high-speed operations in the presence of parameter uncertainties. However, most of these controllers stem from the applications of the existing adaptive control theory, which is traditionally focused on tracking slowly time-varying parameters. In fact, manipulator dynamics have fast transient processes for high-speed operations and load changes are abrupt. These observations motivate the present research to incorporate change detection techniques into self-tuning schemes for tracking abrupt load variations and achieving fast load adaptation. To this end, a robustly global stabilizing controller for a robot model with parametric and non-parametric uncertainies is developed based on the Lyapunov second method, and it is then made adaptive via the self-tuning regulator concept. The two-model approach to online change detection in load is used and the estimation algorithm is reinitialized...

Sequential Detection of Changes in Stochastic Processes

Abstract This report is concerned with the problem of sequential (on-line) detection of abrupt changes in stochastic processes. Many different tasks of control theory, identification theory and signal processing can be solved by making use of such approach. This report presents some results of recent work on development and investigation of the sequential change detection algorithm based upon a likelihood ratio. Both simple (additive) and more complex (spectral) changes are introduced through this general approach. Some typical application of on-line change detection algorithm are also described.