Artificial Intelligence techniques have been effectively used to set standards and check compliance
in hot metal production.
The techniques are used to build relations between masses of historical machine sensor data and the output quality information available from the factory registers and product quality reports produced by quality audit/ factory technical departments. The ranges for individual parameters in which the output quality are compliant are established, and measures are implemented to ensure that the furnace process control equipment work within the established ranges.
The resultant benefits are manifold: 1) Consistent quality in output hot metal; 2) Improved overall productivity; 3) Consequent improved sales figures and bottom line.
Utilization of AI to derive the desired parameter values to control the blast furnace and other iron making operations help in achieving automation and increasing operating efficiency.
This is just one of the many areas and many industries where AI techniques can be used to improve efficiency in factory operations.
Composites are increasingly being used in the automotive and aerospace industries because they
are light (compared to metals), have superior performance at considerably lower weight and can
handle higher temperature ranges. Multi-Fiber systems (Carbon-Glass; Glass-Polypropylene;
Carbon-Nylon; etc.) are used in this process. For the thermoplastic composite to be consistent
quality, it is necessary for the mix of carbon and thermoplastic to be uniform across the cross
section of the composite material.
Early identification of the uniformity and segregating the ones where mixing is not uniform significantly help in improving product quality, thereby reducing costs.
Deep Learning techniques have been successfully used to process images of the cross sections of the fibers to determine both whether the carbon reinforcement has been uniform across the cross section and the percentages of carbon and thermoplastic. The results need to be mapped to the set standards to determine the quality of the composite material.
Microscope images of the cross section of the composites are taken. For a reasonable sample of the microscopic images, the individual fibers are identified and annotated. The model is trained using a large sample of the annotated images. The model learns to identify the different fibers, their texture, color, typical size and shape.
Once the model is built it is capable of numerically expressing the consistency of the mix for any cross sectional microscopic image for the composite on which the model was trained. If any metric for the composite falls below the defined tolerance limits, problems can be identified and remedial action taken.
Image processing and deep learning techniques are being used in many applications across multiple functions and thermoplastic composites is just a niche example. Uses are found in security and surveillance, medical treatment, manufacturing, automotive and gaming, to name just a few.
A petroleum refinery plant uses a number of steam turbines fed by boilers and a series of gas
turbines for its captive power needs for the entire plant.
The weekly and monthly power requirements are calculated based on production targets and the plant has to operate its steam and gas turbines to feed the demand. The management felt that fuel consumption in the turbines was not optimized and looked to AI to achieve cost savings in this area.
Operating data for all the boilers such fuel flow, water flow rates, steam temperature, heat generated, etc. and for all the turbines including fuel flow, gas supplied for the combustion, power generated, exhaust gas composition, etc. for a period of one full year at intervals of 1 minute were used to build the models.
AI models were created for each of the boilers and the turbines. The models suggested the optimal use of fuel and water for the boilers and the optimal fuel flow rate and input gas for the turbines. Significant fuels savings were achieved for the boilers as well as the gas turbines. Once the model for boilers and turbines were in place another model was built to recommend what combination of gas and steam turbines (all with different capacities) should be used to meet the daily/weekly and monthly power requirements.
Most manufacturing plants have implemented preventive maintenance strategies to avoid costly
machine repair and extended down times.
In the absence of a preventive strategy (Run to Failure Maintenance), what is even more costly than expensive machine repair/replacements, is extended down times which impact organizations’ reputation and bottom line significantly.
Traditional preventive maintenance strategies tend to be cautious, leading to additional and sometimes unnecessary expenditure and downtime.
Progressive managements are increasingly looking to predictive maintenance techniques to avoid unnecessary maintenance costs and down time.Instead of correcting equipment failures and breakdowns once they occur, predictive maintenance prevents these problems from ever happening by using intelligent algorithms to predict equipment failure.
With predictive maintenance it is possible to minimize both high repair costs and unnecessarily maintenance. Both past trends and real time data are used offer customers the optimal maintenance choice.
Data used for predictive maintenance comes from a host of sources: manual inspection records, weather information, usage data, manufacturers’ specifications, other environmental data as well as programmable controllers and a plethora of IOT devices.
Predictive maintenance techniques would also decide the spare parts stocking policies, line up the vendors for just-in-time delivery of expensive parts and could tie up with the production schedules.