BMW Canada

Fast, efficient, reliable: Artificial comprehension in BMW Group Production

Munich, Germany. Artificial comprehension (AI) is on
a arise in automotive production. Since 2018, a BMW Group has been
regulating several AI applications in array production. One concentration is
programmed picture recognition: In these processes, artificial
comprehension evaluates member images in ongoing prolongation and
compares them in milliseconds to hundreds of other images of a same
sequence. This way, a AI focus determines deviations from the
customary in genuine time and checks, for instance, either all required
tools have been mounted and either they are mounted in a right place.


The innovative record is fast, arguable and, many importantly,
easy to use. Christian Patron, Head of Innovation, Digitalization and
Data Analytics during BMW Group Production: “Artificial intelligence
offers good potential. It helps us say a high quality
standards and during a same time relieves a people of repeated tasks.”


At a BMW Group, flexible, cost-effective, AI-based applications are
gradually replacing henceforth commissioned camera portals. The
doing is rather simple. A mobile customary camera is all that
is indispensable to take a applicable cinema in production. The AI solution
can be set adult fast too: Employees take cinema of a component
from opposite angles and symbol intensity deviations on a images.
This way, they emanate an picture database in sequence to build a so-called
neural network, that can after weigh a images but human
intervention. Employees do not have to write code; a algorithm does
that probably on a own. At a training stage, that might mean
overnight, a high-performance server calculates a neural network
from around 100 images, and a network immediately starts optimizing.
After a exam run and presumably some adjustments, a reliability
reaches 100%. The training routine is finished and a neural network
can now establish on a possess either or not a member meets a specifications.


Even relocating objects are reliably identified mostly eccentric of
factors such as lighting in a prolongation area or a accurate camera
position. This opens adult a far-reaching operation of intensity applications along
a whole automotive routine chain, including logistics. In many
cases, a AI record relieves employees of repetitive, monotonous
tasks such as checking either a warning triangle is in a right
place in a box or either a windscreen wiper top has been put on.


Artificial comprehension can also perform some-more demanding
investigation tasks

In a final investigation area during a BMW Group’s Dingolfing plant, an
AI focus compares a automobile sequence information with a live picture of
a indication nomination of a newly constructed car. Model designations
and other marker plates such as “xDrive” for four-wheel drive
vehicles as good as all generally authorized combinations are stored in
a picture database. If a live picture and sequence information don’t correspond,
for instance if a nomination is missing, a final investigation team
receives a notification.

Christian Patron: “We rest wholly on a knowledge and expertise
of a employees in these efforts. They can decider best during which
prolongation stairs an AI focus might urge peculiarity and efficiency.
We deliberately keep a setup and doing of such applications
simple. Their operation requires no modernized IT proficiency.”


AI eliminates pseudo-defects

At a press shop, prosaic piece steel tools are incited into
high-precision components for a automobile body. Dust particles or oil
residues that sojourn on a components after combining can simply be
confused with really excellent cracks, that start in singular cases during the
process. Previous camera-based peculiarity control systems during a BMW
Group’s plant in Dingolfing, Germany, spasmodic also noted these
pseudo-defects: deviations from a target, even yet there was no
tangible fault. With a new AI application, these pseudo-defects no
longer start since a neural network can entrance around 100 real
images per underline – i.e. around 100 images of a ideal component,
100 images with dirt particles, another 100 images with oil droplets
on a component, etc. This is quite applicable in a box of
a visually tighten calls that have formerly led to pseudo-defects.

The BMW Group’s Steyr plant and a BMW Group Data Analytics Team are
also successfully operative on expelling pseudo-defects. Presumed
irregularities in torque dimensions in a engine cold exam later
mostly spin out to be insignificant. Before introducing a AI
solution, however, such formula led to formidable primer inspections and
serve exam runs, adult to and including prohibited tests with fuel. The
research program was lerned formed on many available exam runs and
so schooled to heed between tangible and reputed errors.


AI ‘in control’: Integrating synthetic comprehension with
trickery and drudge control systems

The initial intelligent AI control focus during a BMW Group celebrated
a premiere during a BMW Group’s Steyr plant. This focus speeds
adult logistics processes by preventing nonessential transports of empties
on circuit belts. To this end, a containers pass by a camera
station. Using stored picture information noted by employees, a AI
focus recognizes either a enclosure needs to be lashed onto a
pallet or either – in a box of large, fast boxes – no additional
securing is required. If no lashing is required, a AI application
leads a enclosure by a shortest track to a dismissal hire for
a forklift truck. Containers that contingency be additionally secured, on
a other hand, are guided directly to a circuit territory with the
lashing complement and usually afterwards to a dismissal hire located behind.
Previously, all containers had to be ecstatic to a removal
hire for vast containers. From there, a containers that required
additional securing had to be forwarded – and would usually strech the
lashing trickery and finally a scold dismissal hire after taking
this detour.


Besides a focus in Steyr, AI is behind countless other logistics
during a BMW Group. It also supports practical layout
planning, that creates high-resolution 3D scans of buildings and
factories. Artificial comprehension eventually contributes to the
approval of particular objects in a 3D scan, such as containers,
building structures or machines. This allows engineers to remove
particular objects from a 3D indicate in a 3D formulation program and to
cgange these individually, that creates it easier to copy and
know adaptations on a emporium floor.


There is already a graphic trend toward regulating AI applications during the
BMW Group’s plants. The augmenting formation of smart
information analytics
, state-of-the-art dimensions record and AI
opens adult new opportunities in prolongation management. At a physique shop,
for instance, images from a final investigation might uncover that weld
steel has sprayed out during a same welding indicate in several automobile bodies.
Using AI, a control loop can so be sealed and complement control or
upkeep cycles be practiced even faster and some-more efficiently. At
a paint shops, AI and analytics applications offer a intensity to
detect sources of blunder during such an early theatre that errors can hardly
start any more: If no dirt attaches to a automobile physique before portrayal in
a initial place, nothing has to be discriminating off later.