Project Details

Client

QDI

Country

United States

Industry

Robotics

QDI Tank Washing

The Story

A pattern recognition and self-learning system used for washing tanks. QDI had a fleet of 10.000 tanks for chemical bulk transportation. After each transport the inside of the tanks requires to be washed so the next transport chemicals wouldn’t react with the remaining chemical product transported. For this, QDI had a large number of employees which manually assign each tank to a washing facility. The washing facilities for this task are specialized placed, which deals with hazardous materials and the operation is time and money consuming. Each chemical substance transported requires a certain chemical substance to neutralize. In same cases, the chemical substances transported are dangerous and/or flammable so the process needs to be carefully done.
QDI didn’t maintain a DB with what substance needs to be used to neutralize the previous one nor a clear list of specialize washing facilities, this whole knowledge was not fully formalized. The only place in the corporate DB were this information was stored was in the actual washing orders. The manual process of issuing the washing orders didn’t take into account the pending transport orders, due a lack of integrating the transport orders and washing orders, ignoring the possibility that if a tank could be assigned to transport the same chemical substance it will remove the need of washing, resulting in unnecessary cleaning and therefore wasted money.
The software solution proposed was to use an algorithm that will mine the historical data regarding the washing and to detect the patterns regarding what chemical substance will neutralize the transported chemical. Due the fact that the manual entries were not consistent the pattern recognition was employed to make sure it is the same substance. For example, some data entries were “Sulphur acid” while some where “sulph acid” and others were “H2SO4”. The system was able to clear the mess and consolidate a DB based on the historical data. Based on previous historical knowledge, the system was able to make accurate predictions about the washing orders chemical neutralizing substances. Furthermore, based also on the historical data, the system was able to identify the nearest washing facility and to determine also the most cost effective place. Over time, a list of washing facilities was build and consolidated. Another function was to integrate the pending transports with the washing system and, based on the geo location, to identify the nearest tank carrying the same chemical to determine if more feasible to assign it to the transport order rather than washing it.
Due the hazardous nature of the process and the risks involved, the system was not autonomous, the final decisions belonged to a human operator which would had override any decision made by the software AI. The system was self-learning and adapting based also on the feedback of the human operations which over time made that the rate of incorrect proposals decreased dramatically. The need of human data entries was reduced to a minimum and the remaining staff job was to monitor, approve or not the decisions and correct the decisions when necessary.