Automated diagnosis of printer systems using Bayesian networks

A printing diagnosis system comprising:

• a first Bayesian network that handles all errors where a customer does not get output from a printer when attempting to print, and where the customer gets corrupted output;
• a second Bayesian network that handles all errors where the customer gets unexpected output;
• a third Bayesian network that handles all types of errorcodes that can be seen on a control panel of the printer;
• and,
• a fourth Bayesian network that handles miscellaneous erroneous behavior of the printer not covered by the first Bayesian network, the second Bayesian network and the third Bayesian network.

A method for performing knowledge acquisition used to troubleshoot a system, the method comprising the following steps:

• (a) identifying an issue to troubleshoot;
• (b) identifying causes of the issue;
• (c) identifying subcauses of the causes;
• (d) identifying troubleshooting steps;
• (e) matching troubleshooting steps to causes and subcauses;
• (f) estimating probabilities for the causes identified in step (b) and the subcauses identified in step (c);
• (g) estimating probabilities for actions and questions set out in the troubleshooting steps;
• and,
• (h) estimating costs for actions and questions set out in the troubleshooting steps.

A dataflow model of a print job comprising:

• a first Bayesian network model representing how a file can cause erroneous printer output;
• a second Bayesian network model, coupled to the first Bayesian network, representing how an application can cause erroneous printer output;
• a third Bayesian network model, coupled to the second Bayesian network, representing how a printer driver can cause erroneous printer output;
• a fourth Bayesian network model, coupled to the third Bayesian network, representing how a spooler can cause erroneous printer output;
• a fifth Bayesian network model, coupled to the fourth Bayesian network, representing how operating system redirect code can cause erroneous printer output;
• a sixth Bayesian network model, coupled to the fifth Bayesian network, representing how a network driver can cause erroneous printer output;
• a seventh Bayesian network model, coupled to the sixth Bayesian network, representing how a network card can cause erroneous printer output;
• a eighth Bayesian network model, coupled to the seventh Bayesian network, representing how a first network connection can cause erroneous printer output;
• a ninth Bayesian network model, coupled to the eighth Bayesian network, representing how a server can cause erroneous printer output;
• a tenth Bayesian network model, coupled to the ninth Bayesian network, representing how a queue can cause erroneous printer output;
• a eleventh Bayesian network model, coupled t o the tenth Bayesian network, representing how a second network connection can cause erroneous printer output;
• a twelfth Bayesian network model, coupled to the eleventh Bayesian network, representing how printer hardware can cause erroneous printer output;
• a thirteenth Bayesian network model, coupled to the twelfth Bayesian network, representing how the printer hardware output can cause erroneous printer output;
• a fourteenth Bayesian network model, coupled to the fifth Bayesian network, representing how a local parallel cable can cause erroneous printer output;
• a fifteenth Bayesian network model, coupled to the fourteenth Bayesian network, representing how a switchbox can cause erroneous printer output;
• and,
• a sixteenth Bayesian network model, coupled to the fifth Bayesian network, representing how a third network connection can cause erroneous printer output;
• wherein the twelfth Bayesian network model is also coupled to the fifteenth Bayesian network model and the sixteenth Bayesian network model.

A Bayesian network that models dataflow through a system, the Bayesian network comprising:

• an indicator node that has a state that indicates whether there is a failure in dataflow;
• a dataflow components node defining all components of the data flow;
• a plurality of component indicator nodes, a state of each component indicator node representing whether failure in the dataflow is due to an associated component;
• a plurality of flow nodes presenting data flow through the components;
• a plurality of global troubleshooting nodes, each global troubleshooting node being associated with at least one of the plurality of component indicator nodes or at least one of the flow nodes, each global troubleshooting node providing a global troubleshooting step;
• a first plurality of troubleshooting nodes, each troubleshooting node representing a troubleshooting step, each troubleshooting node being coupled to at least one cause node from the plurality of cause nodes, each troubleshooting step suggesting an action to remedy causes represented by any cause nodes to which the troubleshooting node is coupled;
• and,
• a plurality of component network nodes, each of the component network nodes representing a Bayesian network modeling causes of an associated component.

A method for troubleshooting a problem within a system, the method comprising the following steps performed by an automated troubleshooter:

• (a) identifying the problem;
• (b) selecting one of any remaining actions or remaining questions;
• (c) when in step (b) a remaining question is selected, performing the following substeps:
• (c.1) asking a user the selected remaining question,
• (c.2) receiving any answer provided by the user,
• (c.3) if there are no more actions or questions terminating the troubleshooter as unsuccessfully fixing the problem, and
• (c.4) if there are more actions or questions, returning to step (b);
• and,
• (d) when in step (b) a remaining action is selected, performing the following substeps:
• (d.1) performing the selected remaining action,
• (d.2) determining whether the problem is solved,
• (d.3) if performing the selected remaining action solved the problem, terminating the troubleshooter as successfully fixing the problem, and
• (d.4) if performing the selected remaining action did not solve the problem, performing the following substeps:
• (d.4.1) if there are no more actions or questions terminating the troubleshooter as unsuccessfully fixing the problem, and
• (d.4.2) if there are more actions or questions, returning to step (b);
• wherein step (b) includes:
• calculating expected cost of repair (ECR) for performing an optimal action,
• calculating expected cost of observation (ECO) for asking an optimal question, and
• selecting on the basis of ECR and ECO, one of the optimal actions and the optional question;
• and,
• wherein when calculating ECO and ECR, taking into account that when an action that can solve multiple causes fails, probabilities of each of the multiple causes decreases and probabilities of other remaining causes increases accordingly such that a sum of cause probabilities is still 1.

A method for troubleshooting a problem within a system, the method comprising the following steps performed by an automated troubleshooter:

• (a) identifying the problem;
• (b) selecting one of any remaining actions or remaining questions;
• (c) when in step (b) a remaining question is selected, performing the following substeps:
• (c.1) asking a user the selected remaining question,
• (c.2) receiving any answer provided by the user,
• (c.3) if there are no more actions or questions terminating the troubleshooter as unsuccessfully fixing the problem, and
• (c.4) if there are more actions or questions, returning to step (b);
• and,
• (d) when in step (b) a remaining action is selected, performing the following substeps:
• (d.1) performing the selected remaining action,
• (d.2) determining whether the problem is solved,
• (d.3) if performing the selected remaining action solved the problem, terminating the troubleshooter as successfully fixing the problem, and
• (d.4) if performing the selected remaining action did not solve the problem, performing the following substeps:
• (d.4.1) if there are no more actions or questions terminating the troubleshooter as unsuccessfully fixing the problem, and
• (d.4.2) if there are more actions or questions, returning to step (b);
• wherein step (b) includes:
• calculating expected cost of repair (ECR) for performing an optimal action,
• calculating expected cost of observation (ECO) for asking an optimal question,
• selecting on the basis of ECR and ECO, one of the optimal action and the optional question, and
• wherein when calculating ECR, using the following algorithm:
• find first action A1 as action with highest Pi/Ci
• for i=0 to #actions
• recompute probabilities of causes assuming Ai−1=no (fails)
• find action Ai as action with highest p(Ai=yes|A1=no, . . . , Ai−1=no)/Ci.

A method for troubleshooting a problem within a system, the method comprising the following steps performed by an automated troubleshooter:

• (a) identifying the problem;
• (b) selecting one of any remaining actions or remaining questions;
• (c) when in step (b) a remaining question is selected, performing the following substeps:
• (c.1) asking a user the selected remaining question,
• (c.2) receiving any answer provided by the user,
• (c.3) if there are no more actions or questions terminating the troubleshooter as unsuccessfully fixing the problem, and
• (c.4) if there are more actions or questions, returning to step (b);
• and,
• (d) when in step (b) a remaining action is selected, performing the following substeps:
• (d.1) performing the selected remaining action,
• (d.2) determining whether the problem is solved,
• (d.3) if performing the selected remaining action solved the problem, terminating the troubleshooter as successfully fixing the problem, and
• (d.4) if performing the selected remaining action did not solve the problem, performing the following substeps:
• (d.4.1) if there are no more actions or questions terminating the troubleshooter as unsuccessfully fixing the problem, and
• (d.4.2) if there are more actions or questions, returning to step (b);
• wherein step (b) includes:
• calculating expected cost of repair (ECR) for performing an optimal action,
• calculating expected cost of observation (ECO) for asking an optimal question,
• selecting on the basis of ECR and ECO, one of the optimal action and the optional question, and
• wherein when calculating ECO, approximating probabilities of actions solving a problem p(A=yes|Q=s,e) given observation of a question Q in state s utilizing the following equation:
• p ⁡ ( A = yes | Q = s , e ) = ⁢ p ⁡ ( Q = s | A = yes , e ) ↔ p ⁡ ( A = yes | e ) p ⁡ ( Q = s | e ) ⋃ ⁢ p ⁡ ( Q = s | A = yes ) ↔ p ⁡ ( A = yes | e ) p ⁡ ( Q = s | e )

A method for troubleshooting a problem within a system, the method comprising the following steps performed by an automated troubleshooter:

• (a) identifying the problem;
• (b) selecting one of any remaining actions or remaining questions;
• (c) when in step (b) a remaining question is selected, performing the following substeps:
• (c.1) asking a user the selected remaining question,
• (c.2) receiving any answer provided by the user,
• (c.3) if there are no more actions or questions terminating the troubleshooter as unsuccessfully fixing the problem, and
• (c.4) if there are more actions or questions, returning to step (b);
• and,
• (d) when in step (b) a remaining action is selected, performing the following substeps:
• (d.1) performing the selected remaining action,
• (d.2) determining whether the problem is solved,
• (d.3) if performing the selected remaining action solved the problem, terminating the troubleshooter as successfully fixing the problem, and
• (d.4) if performing the selected remaining action did not solve the problem, performing the following substeps:
• (d.4.1) if there are no more actions or questions terminating the troubleshooter as unsuccessfully fixing the problem, and
• (d.4.2) if there are more actions or questions, returning to step (b);
• wherein step (b) includes:
• calculating expected cost of repair (ECR) for performing an optimal action,
• calculating expected cost of observation (ECO) for asking an optimal question,
• selecting on the basis of ECR and ECO, one of the optimal action and the optional question, and
• wherein when calculating ECO, using an expression for a probability that a question Q identifies a cause which can be compared with a probability that an action solves the problem, the expression identifying a cause being expressed as a maximization over causes and states of the question Q, the expression being set out below:
• p Qid ⁢ ⁢ c = max C ⁢ max s ⁢ ⁢ p ⁢ ( C = y | Q = s ) - p ⁢ ( C = y ) 1 - p ⁢ ( C = y ) ↔ p ⁢ ( Q = s )

A method for troubleshooting a problem within a system, the method comprising the following steps performed by an automated troubleshooter:

• (a) identifying the problem;
• (b) selecting one of any remaining actions or remaining questions;
• (c) when in step (b) a remaining question is selected, performing the following substeps:
• (c.1) asking a user the selected remaining question,
• (c.2) receiving any answer provided by the user,
• (c.3) if there are no more actions or questions terminating the troubleshooter as unsuccessfully fixing the problem, and
• (c.4) if there are more actions or questions, returning to step (b);
• and,
• (d) when in step (b) a remaining action is selected, performing the following substeps:
• (d.1) performing the selected remaining action,
• (d.2) determining whether the problem is solved,
• (d.3) if performing the selected remaining action solved the problem, terminating the troubleshooter as successfully fixing the problem, and
• (d.4) if performing the selected remaining action did not solve the problem, performing the following substeps:
• (d.4.1) if there are no more actions or questions terminating the troubleshooter as unsuccessfully fixing the problem, and
• (d.4.2) if there are more actions or questions, returning to step (b);
• wherein in step (b) a question is asked only if it can potentially make a difference in a next selected troubleshooting step.

A method for performing knowledge acquisition to be used by an expert system to troubleshoot a product, the method comprising the following steps:

• (a) identifying an issue to troubleshoot;
• (b) identifying causes of the issue;
• (c) identifying subcauses of the causes;
• (d) identifying troubleshooting steps;
• (e) matching troubleshooting steps to causes and subcauses;
• (f) estimating probabilities for the causes identified in step (b) and the subcauses identified in step (c);
• (g) estimating probabilities for actions and questions set out in the troubleshooting steps;
• and,
• (h) estimating costs for actions and questions set out in the troubleshooting steps.

An expert system comprising:

• a Bayesian network that models a system component causing failure of a system, the Bayesian network including:
• an indicator node that has a state that indicates whether the system component is causing a failure;
• a plurality of cause nodes, coupled to the indicator node, each cause node in the plurality of cause nodes representing a cause of the system component producing a failure;
• and,
• a first plurality of troubleshooting nodes, each troubleshooting node representing a troubleshooting step, each troubleshooting node being coupled to at least one cause node from the plurality of cause nodes, each troubleshooting step suggesting an action to remedy causes represented by any cause nodes to which the troubleshooting node is coupled.