Error Specification, Monitoring and Recovery in Computer Integrated Manufacturing An Analyt

Error Speci?cation,Monitoring and Recovery in Computer Integrated

Manufacturing:An Analytic Approach

A.I.Kokkinaki K.P.Valavanis

Communications Networks Research Group,Robotics and Automation Laboratory, Department of Electronic and Electrical Engineering,The Center for Advanced Computer Studies, Science and Engineering Research Centre,A-CIM Center,

De Montfort University,University of Southwestern Louisiana,

Leicester,LE19BH Lafayette,LA70504

U.K.U.S.A.

aik@https://www.360docs.net/doc/ef1883182.html, kimon@https://www.360docs.net/doc/ef1883182.html,

Abstract

State transitions in Computer Integrated Manufac-turing Systems result either from the execution of the planned actions or from the occurrence of unpredictable events.One key requirement for task planning designed for CIM systems,is the ability to monitor events and respond to them appropriately.Related issues include the speci?ca-tion and classi?cation of possible system faults,errors and failures.This paper proposes an error speci?cation lan-guage,an error monitoring mechanism(EMM)and an er-ror recovery mechanism(ERM).Founded on a mathemati-cal framework,the error recovery mechanism uses the con-cept of severity to alter system characteristics.The paper concludes with a presentation of some examples from an in-tegrated multi-robot system.

1Introduction

The increasing cost and complexity of Computer In-tegrated Manufacturing(CIM)mandates the need for so-phisticated control,monitoring and recovery systems.The research presented in this paper has been motivated by the challenge to derive error speci?cation,monitoring and re-covery methods for CIM systems from the task planning point of view.Task planning methods often assume pre-dictable effects of the executed actions,unlimited resources and perfect information about the domain.These assump-tions,although valid for static environments,do not capture certain characteristics[26]of CIM systems.

In[23]a dynamic task planning method was pre-sented which made the assumption that state transitions in CIM systems may result either from the execution of the planned actions,or from the occurrence of unplanned events.However,as noted in[24],this planning method requires the system to monitor events in its environment and respond appropriately when these events interfere with current plan execution.Monitoring events in a system has also been referred as event tracking[39]or event manage-ment[20].Furthermore,error recovery has been the topic of several studies and various error recovery approaches have been proposed.These may be classi?ed as i)sensor-based reactive mechanisms[1,2,12,21],ii)graph-based methods[4,19,41]and iii)knowledge-based approaches [6,14,17,34,35].Sensor-based mechanisms are tightly

coupled with existing system sensors and actuators.Graph-based methods,apart from being dependent on the exist-ing system model,also result into complicated topologies. Knowledge-based methods do not always meet real time requirements[23].The problem of error recovery in CIM systems becomes increasingly complicated if we also con-sider that CIM systems are typical examples of hybrid sys-tems in which digital components monitor and control con-tinuous physical processes in the analog devices.Although extensive research in computer science and control theory has established well known methods and models for rea-soning about the behaviour of digital and analog systems in isolation,the development of appropriate models for hybrid systems is an active research area[28,32,33].In sum,there is a need to study and develop error recovery approaches that are ef?cient,effective and comply with hybrid systems characteristics.

In this paper concepts and fundamentals related to errors speci?cation,monitoring and recovery have been studied extensively and an analytic approach has been de-rived.First,an error speci?cation language is presented that demonstrates its expressional power through deriva-tion of various classes of system errors.An error monitor-ing mechanism is presented and its output is used as input to the proposed error recovery mechanism.Our approach deviates from conventional error recovery methods by not being dependent on existing system models or components and by adopting continuous mapping of the speci?ed er-rors’severity onto the system parameters.

The rest of the paper is organized as follows:sec-tion2presents an overview of related work in various re-search areas and an informal description of some essential concepts and fundamentals.Section3presents a formal de-scription of the framework.Section4describes an error de-scription language.The error monitoring mechanism and its formal model are presented in section5.The error re-covery mechanism is described in section6.Section7in-cludes some working examples while section8concludes this paper.

2Related Work

Conceptually,an event in CIM systems is similar to a process in qualitative process theory[13].An event,like a process,acts over time to change some parameters of the system.In CIM systems there is also the implicit re-quirement that an event does not belong to the prede?ned set of system actions,which,upon their successful execu-tion,transfer the system from one stage to another.Often, elimination,or prevention of events in CIM systems may not be critical or feasible depending on the required safety features.As noted in[39],events may be instigated by other agents in the system or just occur(i.e.lighting bolt). The Manufacturing Message Speci?cation(MMS)proto-col[20]includes services dedicated to Event Management that support de?nition and management of events or alarms through the use of objects for event condition,event en-rollment and event action.Moreover,the concept of primi-tive and composite events has been examined in Active and Temporal Databases[15,36].

From a task planning point of view,some events may not affect the overall plan execution,whereas some oth-ers may instigate faults,errors,and failures in the system; namely,some events may alter actions’effects and modify the system’s stage.Thus,the overall behaviour of the sys-

tem alternate between plan accomplishment and plan inter-ruption(possibly due to the occurrence of events).Faults in CIM systems may be due to events’occurrence,system deterioration or built-in imperfections[24].A fault can be caused by physical defect,environmental stress or opera-tor’s mistake.Although system sensors may provide some data related to a fault,in general,there is no deterministic speci?cation on the type of the next fault,the component in which it will appear,or the time this will happen.It is assumed,however,that certain characteristics of possible faults are known to the system designers.These include the probability that a speci?c fault will occur and param-eterized information about fault’s effects.Faults may be classi?ed as:i)Permanent faults,which remain in existence until speci?c recovery action is taken,ii)Transient faults, which appear and disappear,usually without causing irre-versible system damage,and iii)Intermittent faults which appear and disappear repeatedly.Finally,it is assumed that the necessary structure to detect and identify faults exists. Thus,fault detection and identi?cation are not issues within the scope of this paper.

An error is the manifestation of a fault,that is,a detected discrepancy between the actual and the expected state of the system.The time between the occurrence of a fault and the resulting error is called error latency.Error latency may be estimated based on the detection capabil-ities of the system.Every identi?ed error has some charac-teristics which may be parameterized.For instance,error parameters include i)the probability of an error happening within a certain period,or dually the mean time before an error is identi?ed ii)the timestamp indicating when the er-ror was identi?ed,iii)the agent(s)affected by the identi-?ed error,iv)the action(or actions,if any)which was af-fected by the error and v)the time required for the error to be handled.The latter parameter is statistically quanti?ed as Mean Time To Repair(MTTR).We assume that error probability is memoryless.In other words,the continuous random variable quantifying the time from some reference start(at which no error has been detected and the system is working according to the speci?cations)until an error is identi?ed is assumed to follow an exponential distribution.

The concepts of extent and span have been derived [24]to quantify the effects of an error in a system.The span of an error refers to the actions which have been af-fected by that speci?c error.Depending on the span of the error,errors may be distinguished into:i)Null point error (no action affected),ii)Single point errors(only one action affected),and iii)Multiple point errors(more than one ac-tions affected).

The extent of an error refers to the agents in the sys-tem that have been affected by that error.Depending on the extent of the error,errors may be classi?ed as:i)Isolated errors(only one agent affected),and ii)Segregated errors (more than one agents affected).It is interesting to note that an isolated error may not be a single point error.To eluci-date this,consider the following case.Assume that an agent has been affected by an error while it was executing an ac-tion and that the affected action has synergetic dependen-cies on some other action(s)being executed in parallel by some other agent(s).In such a case,although only a single agent has been affected by the error,multiple actions have been affected by the error’s occurrence.Similarly,a seg-regated error might be a null point error.For example,an error which affects a number of agents,all of which were

idle at that particular time instant.

An error may be proclaimed to be a failure depending on its severity.That is,a failure is de?ned to be an error with span that contains one or more actions in the current plan.If there are system agents capable of completing the execution of the current plan,then the failure is viewed as a recoverable failure and the system eventually accomplishes the plan.Similarly,if the failure does not permit the sys-tem to complete the current plan,then it is considered an irrecoverable failure and the system is forced to interrupt the current plan and possibly attempt to generate an alter-native plan in order to accomplish the current goal[37].In case of a catastrophic or fatal failure which has affected all system agents,the system will halt.

From the above discussion,it follows that three time intervals pertinent to fault occurrence and error identi?ca-tion can be distinguished.These are shown in Figure1.In each interval there is a class of methods supporting plan achievement.An extended taxonomy of validation and correction methods presented in[27]distinguishes among Fault Avoidance(or Fault Prevention),Fault Tolerance and Error/Failure Recovery.

From the task planning point of view,Fault Avoid-ance,Fault Tolerance and Error Handling are mechanisms to support plan accomplishment.Figure2shows how these mechanisms may be integrated to absorb the effects of faults,errors,and failures in the system.A recovery proce-dure for a failure,or an error may result into continuation of execution of the current plan or in plan interruption.Figure 2was designed with the underlying assumption that only permanent errors require a recovery procedure.Recovery from transient errors is unnecessary and intermittent errors may not be recoverable.

3De?nitions and Formal Framework In this section,we present the formal de?nitions of faults,errors,and failures in a CIM system;an extended analysis of this issue may be found in[25].

D EFINITION3.1The set denotes the set of agents in the system.Every agent is capable of performing(sequentially)a speci?c set of actions over a pe-riod of time.

It is assumed that there is complete knowledge about the agents capabilities and their individual sets of actions. It is also assumed that the agents start their operation at time instant and conclude at time.Also,we use the notation to denote a subset of.

D EFINITION3.2The set denotes the set of actions performed by the system agents.

We use the notation to denote the set of actions that can be performed by the set of agents.

D EFINITION3.3The set denotes the set of the modeled faults in the system.A fault is an unplanned/unanticipated change in the system.Every fault occurrence is associated with the time instant at which the fault came into existence.

D EFINITION3.4The set e/e:system error denotes the set of all possible errors in the system.An error is a dis-crepancy between the actual and the expected state of the system.

Every error is associated with certain characteristics which may be parameterized.An analysis of these param-eterized characteristics follows.

D EFINITION3.5The extent,,of an error,in the system is de?ned in terms of the set of agents which are affected by the occurrence of the error.

affected by

R EMARK3.5.1An error is called isolated,iff .

R EMARK3.5.2An error is called segregated,iff .

D EFINITION3.6The span,,of an error,in the system,is de?ned in terms of the set of actions which are affected by the occurrence of the error.

affected by

R EMARK3.6.1An error is called a null point error,iff.

R EMARK3.6.2An error is called a single point error,iff.

R EMARK3.6.3An error is called a multiple point error,iff.

D EFINITION3.7The severity of an error with respect to an agent is denoted as and is de-?ned as the fraction of the number of the agent’s actions that are included in the error span over the total number of the agent’s actions.

actions while

then,is a recoverable failure.

D EFINITION3.12An irrecoverable failure,is a failure which has affected all system agents capable of exe-cuting the current plan.That is,if the current plan includes actions while

then,is an irrecoverable failure.

D EFINITION3.13A catastrophic,or fatal failure

,is a failure which has affected all system agents.That is,if

then,is a fatal failure.

R EMARK3.13.1For any fatal failure e,

.

D EFINITION3.14Error latency is the amount of time elapsed from the time the fault occurred until the time the error(or failure)was detected.

Based on the presented formal framework and de?ni-tions,an error speci?cation language is derived as follows. 4Error Speci?cation

It is assumed that at initiation time no errors have been identi?ed in the system and at any other particular time instant,some errors(possibly more than one)may have been identi?ed.Furthermore,it is assumed that re-covery of active errors is not instantaneous.Therefore,it is possible that during the recovery procedure of an error another error(s)has been identi?ed which in turn would require the initiation of a recovery procedure-possibly different-than the one to be followed if errors occurred in isolation.

Based on these assumptions,it is possible that at dif-ferent time instances the same fault may result into an error occurring in isolation(primitive error)or be combined with other existing primitive errors and form a composite error in the system.To formalise all possible outcomes,an error speci?cation language is derived.This language must be expressive enough to capture both primitive and composite errors.

4.1Error Speci?cation Language:Constructors

and their Semantics

The error speci?cation language described in this section includes constructors that may be used to specify composite errors.Similar concepts have also been applied in Active Database systems to support event detection[15, 16].An earlier study on this issue was also presented in [24].

D EFINITION4.1The composite error,is called in-clusive disjunction of errors and,where

and,are independent.Inclusive disjunction of errors and occurs when error,or error(or both)occur.

if only has occurred,

if only has occurred,

if both and have occurred.

(1)

if only has occurred,

if only has occurred,

if both and have occurred.

(2)

if only has occurred,

if only has occurred,

if both and have occurred.

(3)

(4)

The probability of disjunction of more than two er-rors can be easily computed by following the generalized principle of inclusion and exclusion.

D EFINITION4.2The composite error,is called conjunction of errors and,where.Con-junction of errors and occurs when both errors and have occurred,regardless of order.

(5)

(6)

(7)

(8)

D EFINITION4.3The composite error,is called se-quence of errors and,where.Sequence of errors and occurs when?rst error and afterwards error have occurred.

(9)

(10)

(11)

(12)

The following three constructors are also referred as ?lters of the errors.These?lters are set depending on how many times a speci?ed error pattern has been identi?ed dur-ing a speci?ed time interval.The default time in-terval for the?lters is from the time the system starts un-til the time the system accomplishes its goals,that is,the interval.However,it is advisable to use the default time interval only when it is absolutely necessary,because the detection of the following?lters is computationally ex-pensive.

D EFINITION4.4The closure?lter of an error denoted by the constructor,that is,IN,is set

at most once,after the?rst detection of even if multiple detections of have occurred during the speci?ed time in-terval.

IN(13)

IN(14)

IN(15) where is the timestamp of the?rst detection of the error .

IN(16)

D EFINITION4.5The times?lter of an error de-noted by the constructor,that is,IN,is set when the error has been detected times during the spec-i?ed time interval.In the degenerate case,this?lter is set every time()the error is detected(alarm sit-uation).

IN(17)

IN(18)

IN(19) where is the timestamp of the detection of the error in the speci?ed time interval.

IN(20)D EFINITION4.6The negation?lter of an error denoted by the constructor,that is,IN,is set when the error has not been detected in the speci?ed time interval.

IN(21)

IN(22)

IN(23)

IN(24)

D EFINITION4.7The null error is set when no errors have been detected during the active ses-sion of the system,that is within the time interval.

(25)

(26)

(27)

(28) It is assumed that.

D EFINITION4.8When the following sequence of errors

IN is identi?ed,where:,IN

,and,then this sequence of errors is called a transient error.

D EFINITION4.9When the following disjunction of errors

IN IN IN IN is identi?ed,where:,IN,

IN IN,,

,and,,,,,

,and then,this disjunction of errors is called

an intermittent error.

D EFINITION4.10An error is called a permanent error,iff it is neither a transient nor an intermittent error.

4.2Error Language Speci?cation:Syntax

In view of the previous discussion,an error speci?ca-

tion language describes how errors,constructors and?lters

of errors may be combined to describe complex erroneous situations.

Grammar de?nes a language for the description of primitive and composite errors.However,is an ambigu-

ous grammar.For example,grammar can have more than one parse trees generating the string.The described error may be the result of i)the conjunction er-

ror followed by,or ii)the sequence con-jugated with the identi?cation of the error.Therefore,it

is imperative to de?ne precedence relations among the er-

ror constructors and?lters.In accordance with the Boolean Algebra,the constructors are evaluated left to right.In the following precedence hierarchy,constructors are presented

in order of decreasing precedence;constructors at the same precedence level are shown at the same line.

With the enforcement of these precedences,the grammar is transformed into grammar.However, grammar contains left recursions;that is the leftmost symbol on the right hand side of the production is the same as the symbol in the left hand side of the production and this may lead the language parser into in?nite looping.The elimination of left recursions from results into the?nal grammar.

GRAMMAR

error error C error(error)?lter fault

C

?lter T error IN[time,time]

T number

number number digit digit

time number:number:number

digit

fault

GRAMMAR

error error temp1temp1

temp1temp1;temp2temp2

temp2temp2temp3temp3

temp3(error)?lter fault

?lter T error IN[time,time]

T number

number number digit digit

time number:number:number

digit

fault

GRAMMAR

error temp1error

error temp1

temp1temp2temp1

temp1temp2

temp2temp3temp2

temp2temp3

temp3(error)?lter fault

?lter T error IN[time,time]

T number

number number digit digit

time number:number:number

digit

fault

After having de?ned an error speci?cation language, a mechanism to monitor the identi?ed errors is introduced. As before,we assume that in general,an occurrence of a fault may introduce one or more errors in the system,while the execution of an error recovery procedure will alleviate or neutralize the effects of recovered error(s).An in-depth analysis of the relative issues and the description of an error mechanism are included in the next section.

5Error Monitoring in CIM systems

To model an error monitoring mechanism for CIM systems,Hybrid Input/Output Automata[29,30]are used.Hybrid I/O Automata are designed as a formal tool to rea-son about hybrid systems.They may be viewed as an ex-tension of Timed Input/Output Automata[28]incorporat-ing increased?exibility on the modeled types of analog be-havior.

The main characteristic of Hybrid I/O automata is that they receive input and react to their environment con-tinuously.Moreover Hybrid I/O automata may compose other Hybrid I/O automata,which can be a very useful fea-ture in modeling complicated asynchronous discrete event systems.A study on the speci?cations and the veri?cation of deceleration manoeuvre in an automated transportation system was presented in[31].

In this section,?rst we include a brief introduction to Hybrid I/O automata theory based on[28],as well as their formal de?nition.Following that,we examine how I/O au-tomata can be used to model an error monitoring mecha-nism.

5.1Model De?nition

D EFINITION5.1In[31],a Hybrid Input/Output Automa-ton,denoted as H,is de?ned as the quintuple:

where:

states(H)is the set of states of the automaton,

start(H)is the nonempty set of start states.It holds that start(H)states(H).

dacts(H)is the set of discrete actions,partitioned into ex-ternal and internal actions.The external actions are further partitioned into input and output actions.

dsteps(H)is a transition relation,dsteps(H)states(H)

dacts(H)states(H)

trajs(H)is the set of trajectories,where each trajectory is a mapping and I is a left-closed interval

of with left endpoint equal to0.The trajectories spec-

ify how states may be associated with times in an interval

of real time.

If,then w is called an I-trajectory of automaton H.If it consists of a single point it is

a trivial trajectory.If w is an I-trajectory then w.ltime is the “last time”of w and it is the supremum of I.Also w.fstate

=w(0)is the?rst state in the trajectory and if I is right-closed then w.lstate is also de?ned to be the last state,that is w.lstate=w(w.ltime).In case I is a closed interval,w.fstate

=s and w.lstate=,then I-trajectory is said to span from state s to state.

If w is an I-trajectory,where I is right closed,is

an-trajectory and w.lstate=.fstate then the concatena-tion,denoted as can be de?ned as the least function such that and

.The concatenation operator can also be ex-tended to a countable sequence of trajectories.

A Hybrid I/O automaton must also specify the fol-lowing two axioms:

A XIOM1(C LOSURE UNDER COUNTABLE CONCATENATION) If is an trajectory,,where all are right-closed,and for all i,then

is in trajs(H).

A XIOM2(C LOSURE UNDER SUBINTERVAL)

If is an I-trajectory,and J is a left-closed subinterval of I then the function given by

is in trajs(H).5.2An Error Monitoring Mechanism

D EFINITION5.2An Error Monitoring Mechanism,de-noted as EMM,is de?ned as a Hybrid Input/Output Au-tomaton:

where:

states(EMM)

start(EMM)

dacts(EMM)are described in terms of input,internal and output actions.

Input:

Internal:none

Output:

Variables:fault-detected,initially set to0and initially set to.

trajs(EMM),and

dsteps(EMM)are de?ned as follows:

+

Precondition:fault-detected=

Effect:

..

.

Precondition:fault-detected=

Effect:

Precondition:.AND.

Effect:

Precondition:.OR.

Effect:

Precondition:.AND..AND.

Effect:

IN[]

Precondition:.AND.

Effect:IN[]

IN[]

Precondition:.AND..AND.

Effect:IN[]

IN[]

Precondition:.OR..OR. Effect:IN

In this section,an error monitoring mechanism, EMM,was proposed based on the model of Hybrid I/O au-tomata theory.It is assumed that there is an underling struc-ture with fault detection capabilities that provides inputs to the error monitoring system.Identi?ed errors can be either primitive or composite in accordance with the error spec-i?cation language presented in section4.Although it was not explicitly presented in this section,error monitoring can also update errors’characteristics like the extent and span.6Error Recovery in CIM Systems

To model the error recovery mechanism in CIM sys-tems we also use Hybrid I/O Automata.The proposed re-covery procedure is fairly straightforward.For every iden-ti?ed error,a recovery procedure is followed for all agents belonging in the error’s extent.The recovery procedure may vary depending on the system.

An Error Recovery Mechanism,(ERM),is de?ned as a Hybrid Input/Output Automaton.The inputs to the error recovery mechanism are the output from error monitoring, namely the identi?ed system errors.The output is the re-covery procedures applied to the agents that are included in the extent of the identi?ed error.The variables used are described in the sequel and there is no internal actions in the model.The formal de?nitions follow.

D EFINITION6.1An Error Recovery Mechanism,denoted as ERM,is de?ned as a Hybrid Input/Output Automaton:

where:

states(ERM)

start(ERM)

dacts are described in terms of input,internal and output actions.

Input:,

Internal:none

Output:

trajs(ERM),and

dsteps are described in the sequence for every identi?ed

error e.

+

Precondition:

Effect:

In this section a model of recovery system is pre-sented.It is assumed that the error recovery mechanism af-fects instantaneously the system agents and there exist no feedback between the recovery mechanism and the recov-ered agent.The upgrade of error recovery with delays and feedback is straightforward but extensive;in[31]there is a detailed analysis on the control with feedback and delays. Also,it is assumed that errors are recovered obeying a FIFO policy;however,we are also investigating policies that re-cover errors based on priorities.

7A Case Study

Examples in this section are based on the con?gu-ration of the Robotics and Automation Laboratory(RAL) within USL.RAL is an Integrated Multi-robot System (IMRS)with three robots having attached vision systems and two conveyor belts arranged as shown in Figure3.The Adept1robot is dedicated to conveyor one and the PUMA robot is dedicated to conveyor two.The Adept3robot can serve either conveyor one or two and it is the only robot ca-pable of picking up heavy items.

Similarly,the global vision system one is dedi-cated to conveyor belt one and the global vision system two is dedicated to conveyor belt two.The local vision systems on Adept1and PUMA can identify objects on conveyor belts one and two respectively,while the local vision system in Adept3can serve both conveyor belts (but only one at a time).Figure4shows all the agents in the system and Figure5demonstrates the system actions.In RAL system:and

It is assumed that in this system faults may occur in a non-deterministic way.However,some faults can be mod-eled in the system.A non exhaustive but representative list of possible faults in the system follows.

+

Conveyor belt one is out of order.

Conveyor belt two is out of order.

Adept1is out of order.

Adept3is out of order.

PUMA is out of order.

The global vision camera on conveyor belt

one failed to identify an incoming object.

The global vision camera on conveyor belt

two failed to identify an incoming object.

The local vision camera on Adept1failed to

identify missing parts of an incoming ob-

ject.

The local vision camera on Adept3failed to

identify missing parts of an incoming ob-

ject.

The local vision camera on PUMA failed to

identify missing parts of an incoming ob-

ject.

Therefore,the set of faults in the system is

.When fault has been detected,then we say that error has been identi?ed,or alternatively error has https://www.360docs.net/doc/ef1883182.html,ing the error speci?cation language we can de?ne the set of errors that are of interest in the system. First,we assume that we are only interested in primitive er-rors.In other words,and the set of identi?ed errors is denoted as.The severity values of the errors with respect to the system agents is presented in Figure7.

The Error Monitoring Mechanism for RAL con?gu-ration,denoted as EMM-RAL,is de?ned as:

(states(EMM-RAL),start(EMM-RAL),dacts(EMM-RAL),dsteps(EMM-RAL),trajs(EMM-RAL))

where:

states(EMM-RAL)

start(EMM-RAL)

dacts(EMM-RAL)are described in terms of input, internal and output actions.

Input:

Internal:none

Output:

Variables:fault-detected,initially set to0and initially set to.

trajs(EMM-RAL),and

dsteps(EMM-RAL)are de?ned as follows:

+

Precondition:fault-detected=

Effect:

..

.

Precondition:fault-detected=

Effect:

With respect to error recovery in RAL we have decided that whenever an error occurs we want to alter the velocity of the involved system agents.The deceleration depends on the severity of the error.More speci?cally, the new velocity of the agent is a function of the current velocity and the severity of the error.Finally,the velocity of those agents that are not affected by the error remains unchanged.Following the error recovery mechanism proposed in section6,the errors recovery mechanism for RAL(ERM-RAL)is de?ned as follows.

(states(ERM-RAL),start(ERM-RAL),dacts(ERM-RAL),dsteps(ERM-RAL),trajs(ERM-RAL))

where:

states(ERM-RAL)

start(ERM-RAL)

dacts are described in terms of input,internal and output actions.

Input:,

Internal:none

Output:,where

trajs(ERM),and

dsteps are described in the sequence for every identi?ed error e.

+

Precondition:

Effect:

This recovery procedure grades gracefully for differ-ent classes of errors.Also,this approach fundamentally differs from the classical Expert Systems paradigm,be-cause there is a continuous mapping of the severity of the error on the behavior of the agents.Following this recovery approach,the recovery procedure for each of the primitive errors is described below.

+

Adept1’s velocity is set to zero.

Adept3’s velocity is reduced by half.

Conveyor belt1is off.

Adept3’s velocity is reduced by half,

PUMA’s velocity is set to zero.

Conveyor belt2is off.

Adept1’s velocity is set to zero.

The speed of Conveyor belt1is reduced by

half.

The local vision system on Adept1is off.

Adept3’s velocity is set to zero.

The speed of both Conveyor belts is set to

zero.

The local vision system on Adept3is off.

PUMA’s velocity is set to zero.

The speed of Conveyor belt2is reduced by

half.

The local vision system on PUMA is off.

Global vision system1is off.

Global vision system2is off.

Adept1’s velocity is reduced by a fourth.

The local vision system on Adept1is off.

Adept3’s velocity is reduced by a fourth.

The local vision system on Adept3is off.

PUMA’s velocity is reduced by a fourth.

The local vision system on PUMA is off.

The recovery system as described above is coherent and stable and also is able to handle composite errors.To demonstrate this point we examine the response of the pro-posed recovery method to a a situation assessment and reac-tive recovery mechanism.Assume that the sequence

has occurred.That is,a fatal error on conveyor belt one has been followed by a fatal error on https://www.360docs.net/doc/ef1883182.html,ing a situ-ation assessment and reactive recovery method when is identi?ed Adept1is set to stop working and Adept3is set to work at a minimum speed.Similarly,when is identi?ed, Adept3is set to work at a maximum speed.Both these reac-tive recovery procedures are valid if the errors occur in iso-lation.However,if has been identi?ed the aforemen-tioned procedure is not necessary,because Adept3needs to

handle objects in just one conveyor belt,namely conveyor belt two.In the proposed approach,the recovery proce-dure will consist of setting the speed of both Adept1and PUMA to zero,while conveyor belt1and Adept3are op-erating sub-optimally,but the whole scheme is a valid one.

In addition to the recovery procedures that are auto-matically generated by the recovery mechanism,it is also interesting to note that it is still possible to specify certain composite errors of interest and override the usual recovery procedure by a speci?cally tailored recovery approach.

In this section we have demonstrated some working examples of the error recovery mechanism as described in section6and we have demonstrated the coherence and ef-?ciency of the proposed approach both for primitive and composite errors.

8Conclusions and Future Work

In Computer Integrated Manufacturing Systems,pre-vention or elimination of errors and failures may not be fea-sible.One key requirement for task planning systems de-signed for CIM systems is the ability to respond to errors and failures in the system.Conventional error recovery ap-proaches do not constitute a widely acceptable solution in CIM systems because it is dif?cult for them to capture all possible states of a dynamic system and their performance may not meet certain requirements.

In this paper,the problem of error/failure handling in CIM systems was studied at large.A formal language for error speci?cation was derived and a classi?cation of errors was presented.Based on the language speci?cations,an er-ror monitoring mechanism(EMM)and an error recovery mechanism(ERM)have been proposed.The error recov-ery mechanism deviates from the conventional approach of rule-based systems by incorporating a continuous mapping of the error’s severity values onto the functionality of the acting agents.

However,there is a number of issues which are still under investigation.In the future we would like to extend the time speci?cation part of the proposed language,elabo-rate the error recovery mechanism to handle errors with pri-orities,extend it to incorporate other system characteristics and investigate whether the proposed approach can be used in other applications.

LIST OF FIGURE CAPTIONS

Figure 1:Classi?cation of fault and error recovery approaches

Figure 2:System faults,errors and failures Figure 3:RAL Con?guration Figure 4:Agents in RAL system Figure 5:Actions in RAL system

Figure 6:Set of actions performed by each agent in RAL system

Figure 7:Severity values of the system errors with respect to system agents

fault

error/failure

TOLERANCE

FAULT error latency

t

ERROR/FAILURE PREVENTION

FAULT RECOVERY

Figure 1:Classi?cation of fault and error recovery ap-proaches

FAULT AVOIDANCE

FAULTS

FAULT TOLERANCE

HANDLING

FAILURES

error/failure recoverable System Imperfections System Deterioration External Events

irrecoverable error/failure

ERROR/FAILURE

ERRORS

Plan Accomplishment Plan Interruption

Figure 2:System faults,errors and failures

Bin1

Bin2

Bin3

Bin4

Conveyor1

Conveyor2

Adept1

Adept3

PUMA

Local Vision on Adept1

Local Vision on Adept3

Local Vision on PUMA

Global Vision 1

SUN SPARC 2

V i d e o C a b l e

Global Vision 2

E t h e r n e t t o R o b o t s C o n t r o l I n t e r f a c e

Ethernet

Figure 3:RAL Con?guration

Track-on-

Pick-from-

Place-in-bin-1

Place-in-bin-2

Place-in-bin-3

Track-on-

Pick-from-

Place-in-bin-4

Move-object-to-

Move-object-to-

Identify-object-on-

Identify-object-on-Figure5:Actions in RAL system

.51110 0111.5

0.5010 00.501

00001 00000 00000 .250000 .250000 0.25000