Abstract:
Machine learning has become an important and interesting eld when addressing com-
plex industrial tasks. In this study, a semi-automatic tire inspection machine for tire
retreading industry which uses machine learning techniques is developed.
Most consumers tend to retread their tires because retreading is an economical and
an eco-friendly method. As a result, retreading industry is now becoming popular in
developed countries as well as developing countries. Initial inspection is the most crucial
activity in the retreading process because tire defect identi cation is performed in this
phase. Failure in identi cation of defects prior to retreading, may cause delamination
of a retreaded tire consequently leading to a disastrous accident.
There exist many advanced machines for initial tire inspection. Widely used method in
the tire retreading industry is nondestructive defect detection based on X-Ray image
processing. This method is very expensive and used by brand new tire manufacturing
companies as well as tire retreading companies in developed countries. In addition
to that, devices with holography and shearography techniques are used to identify
defects which map the tire defects using optical means and are known to be extremely
expensive.
Conventional inspection method is being followed by the Sri Lankan tire retreading
industry as well as other developing countries such as India, Bangladesh etc. due
to its cost e ectiveness in replace of extremely expensive advanced machinery. The
two main tasks performed in conventional inspection method are visual inspection and
hammering test.
Operator carefully observe the worn tire, identify and mark the defects which could
be observed through the naked eye under visual inspection. The identi ed defects can
be classi ed as tire punctures, unwanted metal particles, ply damages, bead damages
and side wall damages. Hammering test is carried out to identify the defects which
are invisible to naked eye such as inner ply separations, ply damages of a tire, inner
canvas damage and small air bubbles in tread area etc. Usually the test is performed
by hammering all over the tire tread area using a brass rod and listening to the resulted
noise di erence by an expertise.
An expert human inspector performs both visual inspection and hammering test. This
manual inspection is often associated with inaccurate results and undetected defects
due to lack of expertise, causing visual fatigue which results in low e ciency and higher
amount of labor costs in local tire retreading industry. Therefore, the main objective
of this study is to eliminate the expert human resource from the initial tire inspection
process and reduce the complexity of the activity.
In this research, visual inspection activity is trained to a model and Faster RCNN
Inception v2 algorithm is used on TensorFlow platform. Image classi cation and tire
defect detection are done with a collection of real-world industrial image data set.
These images were captured using four cameras which were having a capacity of 12
mega pixels each. Basically 220 images were trained using a computer. Hammering
test activity is trained to a model and YouTube-8M algorithm is used with VGGish feature extractor on TensorFlow platform. The sound signal was captured via a normal
USB microphone. The sound signal was analysed using Audacity open source software
and fed as the input to train the model. Unwanted metal particles of the worn tires are
detected using a metal detector. In addition to defect identi cation mechanisms, defect
localisation system is developed using a microcontroller and an encoder. Defects which
are identi ed from image processing or sound signal processing or metal detection,
location of the defect is identi ed with respect to a reference point of tire. This is very
useful for identifying the exact defect location of tire for the operator.
From the obtained results it can be concluded that, the above image classi cation
and sound signal classi cation models provide results with a higher level of accuracy.
Therefore, the expertise labor which is used to perform the initial inspection process
could be replaced by a novice employee. Furthermore the unique and ideal structure
of this developed machine is associated with low maintenance cost. As a result, small
scale companies would be more comfortable with their existing nancial situations when
using this semi-automatic tire inspection machine to enhance their throughput of the
tire retreading process.
Citation:
Perera, H. A. P. B. (2021). Design and development of semi—automatic tire inspection machine [Masters Theses, University of Moratuwa]. University of Moratuwa Institutional Repository. http://dl.lib.uom.lk/handle/123/17592