336
T. Nagahata et al.
degree of deterioration is determined from the angle measured every 180 s. When
the machined arm falls below 0° and reaches a certain angle, it is judged as “failure.”
The deterioration factor and mechanism are registered on the database beforehand. The deterioration factors are specified on the basis of the product models, such
as individual information of parts and mechanisms, environmental conditions such as
“Temperature” and “Humidity,” and operating conditions such as “Rotational speed”
and “Loads.” In this way, deterioration factors such as “Fast contact”, “Multiple
contacts”, and “Heavy load” may be identified. Then, the deterioration mechanism
caused by the identified deterioration factor is searched from the database. In this
case, “Motion of friction” is identified from “Fast contact”, “Multiple contacts,” and
“Heavy load,” whereas “Slide of axis” is identified from “Heavy load” and “Multiple contacts”. The deterioration mechanism caused by the deterioration factors,
including those generated by the identified deterioration mechanism, is searched
again from the database. This process is repeated until no new deterioration factors
appear, and the deterioration process is generated by linking the determined multiple
deterioration mechanisms. The algorithm for creating the deterioration process is
shown in Fig. 23.11.
The factors that cause the deterioration are regarded as input events, and the
factors that are caused by the deterioration are regarded as resultant events. Based on
this relationship, a deterioration simulation is performed to calculate the conditional
probability between events. The judgement on the occurrence of the resultant event
is carried out based on the judgment function registered in the deterioration database.
By assigning the conditional probability calculated by the deterioration simulation to the deterioration process, the causal relation model is created. The Bayesian
estimation is performed using this causal model to estimate the failure probability.
The estimated failure probability is incorporated into the path probability, as
shown in Fig. 23.9. Then, the expectation value in each route is calculated by multiplying this possibility with the value of the stage in the route, and the appropriate
maintenance action is decided.
23.9 Discussion
More research is required for practical application of the system proposed in this
paper.
First issue is how to identify the causes of deterioration for a product. We need
some measures to extract a specific event efficiently from all the existing known
events related to deterioration based on product model information. Other issues
include how to obtain environmental conditions and usage as well as how to prepare
deterioration database.
We need to study how to deal with the situations when enough clear information
is not available for the creation of causal relations, such as the prior probabilities of
inputs and the judgment functions. It is also a research issue how to determine the
cause of deterioration when no related events are observable.
T. Nagahata et al.
degree of deterioration is determined from the angle measured every 180 s. When
the machined arm falls below 0° and reaches a certain angle, it is judged as “failure.”
The deterioration factor and mechanism are registered on the database beforehand. The deterioration factors are specified on the basis of the product models, such
as individual information of parts and mechanisms, environmental conditions such as
“Temperature” and “Humidity,” and operating conditions such as “Rotational speed”
and “Loads.” In this way, deterioration factors such as “Fast contact”, “Multiple
contacts”, and “Heavy load” may be identified. Then, the deterioration mechanism
caused by the identified deterioration factor is searched from the database. In this
case, “Motion of friction” is identified from “Fast contact”, “Multiple contacts,” and
“Heavy load,” whereas “Slide of axis” is identified from “Heavy load” and “Multiple contacts”. The deterioration mechanism caused by the deterioration factors,
including those generated by the identified deterioration mechanism, is searched
again from the database. This process is repeated until no new deterioration factors
appear, and the deterioration process is generated by linking the determined multiple
deterioration mechanisms. The algorithm for creating the deterioration process is
shown in Fig. 23.11.
The factors that cause the deterioration are regarded as input events, and the
factors that are caused by the deterioration are regarded as resultant events. Based on
this relationship, a deterioration simulation is performed to calculate the conditional
probability between events. The judgement on the occurrence of the resultant event
is carried out based on the judgment function registered in the deterioration database.
By assigning the conditional probability calculated by the deterioration simulation to the deterioration process, the causal relation model is created. The Bayesian
estimation is performed using this causal model to estimate the failure probability.
The estimated failure probability is incorporated into the path probability, as
shown in Fig. 23.9. Then, the expectation value in each route is calculated by multiplying this possibility with the value of the stage in the route, and the appropriate
maintenance action is decided.
23.9 Discussion
More research is required for practical application of the system proposed in this
paper.
First issue is how to identify the causes of deterioration for a product. We need
some measures to extract a specific event efficiently from all the existing known
events related to deterioration based on product model information. Other issues
include how to obtain environmental conditions and usage as well as how to prepare
deterioration database.
We need to study how to deal with the situations when enough clear information
is not available for the creation of causal relations, such as the prior probabilities of
inputs and the judgment functions. It is also a research issue how to determine the
cause of deterioration when no related events are observable.