Anda di halaman 1dari 4

technology

Summary Decision-support
system for bearing
The SKF Panloc bearing unit
offers customers a simple way to
adjust internal clearance/pre-
load,high rigidity,low operating failure mode analysis
temperature and easy replace-
ment of the bearing unit.In terms Gaining insight and information from rolling bearing
of overall costs,this unit is an
damage and failures is of strategic importance for SKF
optimal bearing unit that can be
adjusted to specific applications
and its customers. The knowledge collected on bearing
and customer needs.In addition, damage is accessible for SKF engineers as a web-enabled
the advantages of the SKF Panloc
decision-support system called SKF Bearing Inspector.
ensure increased operating relia-
by GERARD SCHRAM , SKF Reliability Systems, and BAS VAN DER VORST , SKF Engineering &
bility for the customer’s system.
Research Center B.V., Nieuwegein, the Netherlands
Research and development
within SKF in terms of this bearing

T
he decision-support system SKF Bear-
unit are still in progress.
ing Inspector is aimed at offering
Investigations are pointed in increased speed, consistency and
higher quality in the bearing decision-
the following direction:
making process. It should help to prevent
■ seal and lubrication -> lifetime bearing damage or failure from reoccurring.
As with any knowledge-based computer
lubrication Heavy wear on the outer ring of a
system, SKF Bearing Inspector gathers all the
cylindrical roller bearing oper-
■ increasing the basic load rating relevant information and experience avail-
ating in an electric motor of a
able about rolling bearing damage – from
-> greater service life and rigidity paper winder in the reel section
basic principles to practical engineering
of a tissue paper machine.
■ a new concept for rubber results.
Current knowledge-based systems have
cylinder bearings in printing
benefited from the experience of expert
machines. systems developed in the 1980s, although
these suffered major flaws in aspects of
Over the course of this year,addi-
reasoning capacity and computer power.
tional and more detailed results These systems were often structured as
decision trees that led from symptoms to
in this regard will be
possible causes. Causal relations between
available. symptoms and possible reasons do not Small pitting is observed after
exist in reality and can easily lead to wrong further microscopic inspection of
conclusions. This is simply because the the raceway.It also shows a small
reasons (e.g., wrong bearing mounting) layer of rehardened material,due
result in the damage symptoms (e.g., fret- to local high temperature.
ting signs), and not the other way around. A
modeling of a relationship from causes to
symptoms where uncertainty is attached to
Read more at
“possible failure states” fits much better with
http://evolution.skf.com the physical phenomena that occur during
bearing service life. With the aid of

1/03 evolution.skf.com EVOLUTION I 25


state-of-the-art computational intelligence Analysis of bearing damage and failure is and verified or rejected. For example, the
techniques, this approach has been followed principally a diagnostic task. Imagine a doctor starts investigating the most probable
for the development of the program. patient visiting his doctor with a specific diseases by conducting specific medical
complaint. The doctor first questions the tests (blood pressure, heart rate, etc.).
Knowledge system patient about specific body and life-style With a probabilistic network, the two-
Within a knowledge system, one generally parameters such as weight, smoking, etc. step reasoning is implemented by forward
distinguishes between modeling the know- (conditions). Based on that, the doctor and backward probability calculations.
ledge with a certain knowledge representa- makes hypotheses about likely diseases
tion and the reasoning principle, in order to (failure modes). The doctor verifies or Probabilistic network
derive problem-solving capacity. Regarding rejects these hypotheses through further The probabilistic network is a visual net-
knowledge representation, several forms questioning and inspection of the patient work in which nodes are connected by
exist, such as: (symptoms). causal relationships, and probability calcu-
The process of a damage or failure analy- lations are applied. The network for bearing
■ Cases: Many bearing failure experiences sis is similar to the doctor’s approach. In a failure analysis has four node categories:
can be found in case examples. correct diagnosis, there are two reasoning conditions, internal mechanisms, failure
Unfortunately, many practical cases are steps: modes and observed symptoms. Conditions
not well documented, and no uniformity 1. Hypotheses generation is where possible represent the conditions from and under
regarding the documented parameters failure hypotheses are generated based on which the bearing operates. Examples are
or failure mode conclusions exists. data. For example, the doctor starts asking speeds, bearing type, load, temperature,
Example cases can, however, be used to questions to get an idea (hypothesis) of installation details, environmental factors,
model or verify other knowledge what could be wrong; etc. Internal mechanisms represent the
representations. 2.Verifying or rejecting hypotheses. One by physical phenomena that happen during
one, the generated hypotheses are investigated operation, such as lubrication, film disruption,
■ Rules: It is possible to generalize if-then
rules between observed symptoms and
possible causes. However, this is not
appropriate because different causes
can have similar effects that appear as
similar symptoms.
ROLLING BEARING FAILURE MODES
Fatigue Subsurface-initiated fatigue
■ Artificial Neural Networks:
Surface-initiated fatigue (surface distress)
Mathematical relationships between
Wear Abrasive wear
symptoms and causes can be derived by Adhesive wear
using example failure cases. However, Corrosion Moisture corrosion
there are not sufficient number of Frictional corrosion Fretting corrosion
discriminating cases to do this. False brinelling
Furthermore, system users require Electrical erosion Excessive voltage
additional explanations rather than Current leakage
“black box” artificial neural network Plastic deformation Overload
relationships that do not carry such Indentation Indentation from debris
explanations. Indentation by handling
Fracture Forced fracture
Fatigue fracture
■ Probabilistic Networks: It is possible to
Thermal cracking
derive visual networks in which nodes
are connected by causal relationships, Table 1.
based on bearing failure theory and
experience. Furthermore, probabilities CONDITIONAL PROBABILITY THAT SLIDING CONTACT IS TRUE OR FALSE,
are assigned indicating the weakness or GIVEN ACCELERATION IS TRUE OR FALSE
strength of those relationships. By intro- P(sliding contact | accelerations) Sliding contact = TRUE Sliding contact = FALSE
ducing correct causality from conditions Accelerations = TRUE 0.6 0.4
to observations, this knowledge repre- Accelerations = FALSE 0.2 0.8
sentation best fits the bearing failure
diagnosis problem. Table 2.

26 I EVOLUTION evolution.skf.com 1/03


technology

sliding contact, etc. Failure modes repre-


sent the types of failure, such as subsurface-
initiated fatigue and fretting corrosion. In
table 1, the various failure modes are
listed. Observed symptoms represent the
observable phenomena inside and outside
the bearing, including discoloration, Example: step 1: Application conditions are filled by loading the electric motor winder data
spalling, rust, etc. among other bearing type,coating,speeds,etc.Detailed information and
About 150 nodes are connected by causal examples are provided under the information button.
relations between conditions of the bearing
application, hidden mechanisms, (physical)
failure modes and observed symptoms.
In the modeling of the network, various
sources of information were used. Apart
from defining the nodes, the causal relations
and probabilities, explanation texts (for
each node) including examples and pictures
are developed. In total, about 250 pictures
have been included in the system. Example: step 2: Inspection on symptoms for current leakage failure mode.After inspection
and enlargement of the runway surface,small pitting is confirmed.Several examples are pro-
Calculation step 1
vided under the information button.
Hypothesis generation: Once the network
is modeled, the reasoning process can start.
The initial nodes (no inputs) have two or
more states. Each state is assigned a “prior”
probability between 0 and 1, with the total
over the states being 1. For example:
■ P (accelerations = TRUE) = 0.05
■ P (accelerations = FALSE) = 0.95
In the system’s user interface, the user can
state that acceleration is true. This will
change the above probabilities into 1.0 and
0.0, respectively. In the network, condi-
tional probability tables are defined (table 2).
Example:Final diagnosis:results based on the provided application conditions (step 1) and bear-
When nodes have more states (more than
ing system inspections (step 2).Both the probabilities of the most relevant failure modes and
only true and false) or when they have more
related internal mechanisms are listed.The results can be printed out as MS Word or HTML
input relations, the tables grow. With the
report.
conditional probability tables, the probabil-
ities of the other nodes can be calculated by
the formula:
■ P(B) = Σ P(B|Ai) · P(Ai), for all i
with P(B|Ai) being the conditional
probability given the condition Ai. In the be determined and ranked. This is the bearing to simple or complex laboratory
example: failure hypotheses generation. Notice that tests. For this purpose, one first has to
■ P(sliding contact = TRUE) = 0.6 · 0.05 + uncertainties are attached to the node states explain how probabilities of observations
0.2 · 0.95 = 0.22 rather than to if-then rules in a classical will influence the probabilities of the failure
■ P(sliding contact = FALSE) = 0.4 · 0.05 expert system. hypotheses. As this is causally different,
+ 0.8 · 0.95 = 0.78 one has to reason backwards. Without
In this way, all the probabilities of the Calculation step 2 going into detail, the heart of this reasoning
nodes are calculated, given the prior prob- Verification or rejection by inversion: After lies in the formula:
abilities of the start nodes. By considering hypotheses are generated, we have to verify ■ P(B|C) = P (C|B) · P (B) / P(C)
the application conditions as the start nodes, or reject them by investigating the bearing. This states that the belief in hypothesis B
the probabilities of the failure modes can This ranges from visual inspection of the obtaining evidence C can be computed by

1/03 evolution.skf.com EVOLUTION I 27


multiplying our previous belief P(B) by the of a bearing damage investigation. One results in a list of damage symptoms most
conditional probability P(C|B) and that C is exampleis of an electric motor in a paper mill. relevant to the selected failure mode. The
true if B is true. The conditional probabil- In this case, an electrically insulated cylindrical bearing is first inspected for false brinelling.
ities P(C|B) are modeled in the network as roller bearing NU 322 ECM/C3VL024 is Because no shallow depressions are found
causal relation, P(B) comes from forward used in an electric motor of a paper winder that can verify false brinelling, this failure
reasoning (step 1), and P(C) are set during in the reel section of a tissue paper machine. mode is rejected. The analysis is continued
the bearing investigations. These are the The electric motor speed is variable (400 with inspecting of symptoms of adhesive
observations that serve as evidence for the VAC with frequency converter) and running wear. None of the symptoms related to
failure hypotheses. P(B|C) is called the between 1000 and 1500 min-1. After only a adhesive wear are found either. Finally, by
posterior probability, while P(B) is the prior month of operation however, heavy wear inspecting electrical current leakage symp-
probability. was observed on the inner and outer rings. toms, the presence of small pitting is found
Instead of investigating all possible Loading the example case in SKF Bearing- after magnification of the raceway surface.
observations and non-filled-in conditions, Inspector sets all known application condi- This verified the current leakage failure
the most relevant ones are suggested, depen- tions (step 1). The first hypothesis of possi- mode. Subsequently, the customer indeed
dent upon the failure hypothesis (or internal ble failure modes is calculated based on discovered an earthing problem in the
mechanisms) that need to be investigated. these application conditions. At this point winder construction causing the electrical
In other words, these are the application in the analysis, SKF Bearing Inspector gives current leakage.
conditions or observations that have the a high likelihood of false brinelling, adhesive
most discriminating effect on the failure wear and current leakage. At first sight Conclusions
hypothesis. The discriminating effect is current leakage and false brinelling seem SKF Bearing Inspector meets the need for a fast,
determined by a mathematical measure. unlikely because the machine uses insulated more consistent, high-quality decision-
For all possible not-filled-in conditions or bearings and all machines are properly making process for bearing damage and
observations, this measure is scaled supported with rubber pads. failure investigations. This web-enabled system
between 0 and 100. An example is given in The user then has to perform the second is available for SKF engineers to support
the illustrations. Eventually, by investigat- step of the analysis by inspecting the bear- customers in bearing damage and failure
ing the application conditions and obser- ing on failure symptoms. Clicking “inspect” investigations. ❒
vations, the likelihood of the failure
hypotheses and internal mechanisms are
determined and ranked. These then form
the conclusions of the bearing damage
analysis.
The system is further extended with

Summary
various functions that can help the user.
A simple file with user instructions is
provided for getting started. The system
offers translation of key terms into English,
German, French and Swedish. Session SKF has put its bearing expertise and knowledge into a decision-support system
data control is available for session data available for SKF engineers.The system has been
storage and retrieval. Also, through a file
marked “Typical Examples,” users can be developed to meet the need for a fast,more consistent,high-quality
guided through the application of the decision-making process for bearing damage and failure investigations.
program. For convenience, an extensive
report can be generated in MS Word or The system draws its experience from the wealth of information available
HTML format, including the relevant con- from experts,customers,practical research and published documentation on
ditions, observations and failure mode
probabilities. bearing performance and failure modes.The system overcomes the short-

comings of previous expert systems and incorporates improved decision-making


Practical example
The SKF Bearing Inspector contains several processes that help identify the true causes of bearing failures.
common bearing damage cases located under
“Typical Cases.” These can be used as
training material to demonstrate how the
SKF Bearing Inspector supports the analysis

28 I EVOLUTION evolution.skf.com 1/03

Anda mungkin juga menyukai