Skip to main content

Analysis weld pool on-line adaptive weld process control hazardous environments & manufacturing applications using neural networks


The main benefit of NEUROWELD comes from the reduction of the number of weld defects and thus improving the security and reducing the repair cost as well as the costs of welding procedures development in laboratory by optimising the number of tests to be performed.
So, such a development should increase the effectiveness of European end-users during their work for developing a welding process and during their welding operations. It will have an impact on the price of welded constructions, on the integrity of welded structures and will permit to progress in the way of the fully automatic welding process in hazardous environment and manufacturing applications.
The following steps have been carried out under this EC project in order to successfully develop a NEUROWELD prototype which has been validated in industrial conditions using on-site production conditions ;
- Development of an sensor capable to look the weld pool and measure the groove geometry (gap, Hilo) and the data processing algorithms.
- Development of the Neural Network Controller used in auto-adaptive configuration,.
- Integration of the Neural Network control system with a Fronius' automatic welding machine.
- Teaching of the Neural Network and validation of the methodology with TIG, MIG/MAG and TIME twin on stainless steel, carbon steel and Aluminium.
- Development of an appropriate methodology for qualifying welding procedure and systems, which use this kind of auto-adaptive control.

Further to different tests on different metals using the complete NEUROWELD system for TIG, MIG/MAG as for T.I.M.E twin, the validations have been demonstrated that the concept of welding control by Neural Network is able to work correctly and to offer the best results for good and repetitive welds.
All welding qualification results, especially with MAG on carbon steel and MIG on Aluminium, have shown that the Neural Network prototype is able to manage welding conditions for a repetitive quality of welding with several welding processes. It depends only on the initial trials in order to train the Neural Network and constitute the strongest data as a high powerful memory for the future enfaced welding situations.
Concerning the impact on welding codes, a new approach can consist in defining a better range of approval, through a new concept. This concept consists in testing, during the qualification of the welding process and on-line tests, the mini-maxi gap values, eventual mini-maxi hilo values, in order to qualify that welding process with all guarantees of high welding quality, even, if the conditions of welding are very poor due to bad penetration of pieces or if, due to risk environment, good penetration would take too more time before welding.
This new way for the qualification of auto-adaptive welding process can offer the manufacturer a reduction of costs, if the preparation of pieces to be welded can be made without the same quality level as for classical automatic welding.
Further to the completion of the NEUROWELD programme, the ALSTOM-LHB partner has planned to equip its production site in Salgzitter with a first NEUROWELD equipment in order to train its operators to use it under production welding conditions of train roof.
The most dramatic problem encountered when using
automatic welding in all situations associated with
problems of alignment and cast variations between parts
to be welded is major penetration defects, especially for
the critical root pass. Whereas a skilled welder may
easily modify the welding parameters (welding speed,
torch oscillation, etc.) in real time as a function of
his perception of the weld pool, automatic welding is
only possible if the machining and matching tolerances
are accurate. These tolerances are difficult or costly to
achieve, especially in the case of remotely controlled
operation under hostile environments. In the
manufacturing industries, off:line programming of robots
for small production batches is a much time consuming
phase of the production (up to 8 and 12 hours of factory
time plus cost of welding tes runs on components). The
objective of this project is to extend the automation of
the TIG welding process by incorporating a weld pool
measurement sensor and adaptive weld parameter generator
to achieve welds with no penetration defect. The present
state of the art offers no technical solutions meeting
the project objectives; a thorough survey has been made
for more than 4 years, which indicated that the attempts
??de so far both within Europe and outside have either
stagnated at the laboratory stage or enabled only the
joint fit up preparation to be accommodated; identical
casts have been successfully welded but, when attempting
to join parts of same grade but of different cast
origins, penetration defects develop.

The technical objectives are:
1. Automatically generate in real time welding parameters
such as to guarantee complete and accurate penetration of
the root pass with different casts and non perfect fit
2. Eliminate the need for extensive preparatory
laboratory coupon welding runs; This is the most drastic
drawback of knowledge based algorithms which require up
to 3 months of experimental welding work to construct a
data base for one single configuration (grade, thickness,
bevel geometry, position). This can be reduced by a
factor of 20, through the outcome of this research.
3. The system shall equally be capable of automatically
generating torch movements and parameters for the filling
passes such that the bead position, the bead shape and
the fusion of the walls of the groove comply with quality

The major research tasks are:

O Development of a weld pool properties monitoring sensor
based on infra red light analysis combined w ith
vibration response of the molten metal. In order to make
the system to operate in real time, the image processing
will require to be completed in 40 ms, which is 25 times
higher speed than at present. This will be achieved by
the use of parallel processing systems to segment the
image Either an INMOS transputer system or TEXAS C 40
system will be considered as the processing hardware.
O Study of the correlations between the weld pool
properties and the penetration, O Integration of the
previously developed system for on line laser measurement
of the joint fit up geometry,
O Development of a neural network which can be trained
with the original welding data base and then updated with
the information obtained from the sensors,
O Execution of a selected programme of TIG and MIG test
welds on carbon steels, austenitic and duplex S.S.
aluminium and titanium alloy to verily that the system is
effectively capable of automatically producing welds with
the specified penetration using specimens of different
casts, for cach grade.

The consortium comprises of: 7 partners from 5 countries
of EU, 4 industrial end users, out of which 2 are SMEs;

-> ARSOPI (PT), which is an SME specialised in
fabrication of food, process, agro and petrochemical
industry welded components in different materials:
carbon,low alloy. austenitic and duplex stainless
steels. aluminium and titanium.

-> The UNIVERSITY OF LIVERPOOL (UK), having expertise in
parallel processing (transputers), neural network based
machine control systems, high speed image processing and
weld pool properties sensors.

-> COMEX NUCLEAIRE (FR), having the role of both an end
user and welding systems developer, is an SME (Co

-> LINKE HOFMANN BUSCH (DE), an end user manut`acturing
railways vehicles, metro coaches, etc. (Transport
Industry Sector)

-> FRONIUS (AT). end user. a manufacturer of TIG and MIG
MAG welding machines and robots marketed woridwide.

-> ISQ (PT). cxpert in welding metallurgy and robotic
welding with valuable experience and a National
Accredited Body for W elding, Inspection, Certification
and Qualifications.
-> BUREAU VERITAS (FR). will play a key role to make the
project results effectively exploitable by the EU
industry from a normative (standards) point of view.

Invito a presentare proposte

Data not available

Meccanismo di finanziamento

CSC - Cost-sharing contracts



Partecipanti (6)

ARSOPI - Industrias Metalurgicas Arlindo S.Pinho SA
3731 Vale De Cambra

Mostra sulla mappa

Bureau Veritas S.A.
17 Bis,place Des Reflets 17 Bis
92400 Pari Courbevoie

Mostra sulla mappa

Fronius Schweißmachinen KG Austria
15-17,gewerbestraße 15-17
8754 Thalheim

Mostra sulla mappa

Instituto de Soldadura e Qualidade
Tagus Park
2781 Oeiras

Mostra sulla mappa

Linke-Hofmann-Busch GmbH

38233 Salzgitter

Mostra sulla mappa

United Kingdom
Brownlow Hill
L69 3GJ Liverpool

Mostra sulla mappa