You are on page 1of 13

U/P 375

Vehicle Failure Detection


& Business Application

12/10/2019
NVH INTRODUCTION
 Background: Noise, vibration and harness (NVH) is used to study
noise and vibration vehicle characteristics. Every type of vehicle
has its unique vibration/noise performance. NVH correlates with
vehicle failure modes and is used as a detection method in Failure
mode and effects analysis (DFMEA). If out of spec noise related to
severity 9 and 10 can be detected successfully, and instant waring
can be provided to drivers, crash rates will be significantly
reduced. Lower detection values in DFMEA can offset investment in
failure prevention per government requirements, reducing
engineering and material costs.

2
NVH INTRODUCTION
 Issues in NVH: Traditionally, NVH data is collected and processed
by NVH department. It is easy to correlates failure with noise when
there is only one type of failure since every component or defection
has its own natural frequency. When multiple components fail at
the same time, it is difficult to locate the true root cause of the
problem. Swap test and Fast X projects are required to eliminate
each potential root cause, costing tremendous amount of lab
 resource and man
Project Scope: Usepower.
artificial intelligence (AI) Keras deep learning
library to link between NVH data and vehicle transmission gear box
failure modes.

3
Testing Setups
 Experiment Method in Data Collection
- 1. Locate the vehicle with bad acoustic performance
- 2. Objectively verify that the noise/vibration is repeatable
- 3. Prepare research grade micro phone and put stick them to suspected areas
of the gear box
- 4. Record sound maps of the noise in each area
- 5. Tear down the gear box and send it to running noise bench for testing
- 6. Send out an input signal to gear box and generate noise frequency vs
magnitude plot
- 7. Use harmonic frequency calculation method to locate identify potential
study components
- 8. Use swap test to eliminates other assumptions
- 9. Confirm the root cause by reassemble the WOW component into well
functioned gear box which has all BOB components
- 10. Generate enough data set for deep learning analysis
- 11. Use deep learning model to predict failure mode when out of spec noise
4
appears
Data Processing
 Build Acoustic Sound Recognition:
- 1. Collect gear box data from micro phone
- 2. Feature extraction from sound map use encoding

5
Data Processing
 Build Acoustic Sound Recognition:
- 3. Data modeling (use 2 dimensional vector numpy array to characterize noise
data) through encoding model

 Numpy Array Structure in a matrix model


[ 0, 1.5, 3.7, 21, 47, 56, Time
0.1, 0.3, 0.4, 0.8, 0.9, Amplitude
120, 130, 150, 170, 210] Occurrence

6
Data Processing
 Build Failure Mode Recognition:
- 1. Collect frequency map from failed component from the failed gear box
- 2. Feature extraction from vibration plot
- 3. Data modeling through encoding model

 Numpy Array Structure in a matrix model


[ 1, 5, 7, 12, 15, 11, Time
0.1, 0.2, 0.7, 0.8, 0.9, Amplitude
170, 110, 155, 179, 250] Occurrence

 Repeat the process for other components

7
Deep Learning Training Data
 Build Training Data Set
- Input: Encoded noise data from multiple locations
- Output: Encoded frequency data from components

 Separate 20% of data to testing

8
DNN and Deep Learning
 Use data to train the model
- Testing the different numbers of layers and neurons

9
DNN and Deep Learning
 Outcome
- After the DNN is trained by the data set, now we can predict which component
failed based on noise collected from multiple location of the gear box.

10
Ethics
 Privacy
 Data-collection
 Consent from drivers
 Only collect as much data is necessary

 the responsibility of AI practitioners


Guide the data well
Be aware that it is extremely hard to anonymize data, especially with powerful social network analysis
techniques.
 Data-Usage

 Only release as much data as necessary

11
Ethics

 Type I Error
 Overstatement of defects- Unnecessary Cost to car maker and insurance
company
 Type II Error
 Fail to detect defects- Accidents/ Law-suits

 Replacement Threat
 Firing Inspectors
 Conflict of Law
 Different regulations among jurisdictions

12
Ethics
 ML easily reinforces biases/Ways to enhance accountability
 •Responsibility
 • Make available externally visible avenues of redress for adverse individual or
societal effects of an algorithmic decision system, and designate an internal role
for the person who is responsible for the timely remedy of such issues.
 • Explainability
Ensure that algorithmic decisions as well as any data driving those decisions
can be explained to Drivers, Automaker, Insurance Company, and other
stakeholders in non-technical terms.
 • Accuracy
 Identify, log, and articulate sources of error and uncertainty throughout the
algorithm and its data sources so that expected and worst case implications can
be understood and inform mitigation procedures.
 • Auditability
 • Enable interested third parties to probe, understand, and review the behavior of
the algorithm through disclosure of information that enables monitoring,
checking, or criticism.
13

You might also like