The Container Terminal industry has started to recognise that improving efficiency requires a new paradigm. That automation cannot be achieved by over elaborating existing systems and equipment only serves to embed inefficient processes even more. The adoption of more ‘factory like” layouts, Rail Mounted Gantry Cranes [RMGs] and all-electric deigns is accelerating. But these attempts largely fall short of full automation; at least as understood in heavy industry. There are two emerging models of the automated container terminal. The first is typified by an arrangement of the Container Yard [CY] perpendicular to the quay with numerous long narrow container stacks. Each stack is generally served by two RMGs. This arrangement has been called the “End-Loaded” design. The second model is a parallel arrangement of the CYs served by a smaller number of larger RMGs. This arrangement has been called the “Side-Loaded” design. Of the latter design a distinguishing feature is the method used to move containers to and from the stacks and the quay. In this paper we elaborate on the methods of operation of each approach and provide CAPEX and OPEX reference points for comparison of both designs and their operating methods. The perpendicular or ‘End-Loaded’ design has be implemented by HHLA at Altenwerder Germany [CTA]; by APMT at Portsmouth [now owned by Virginia Ports Authority], and at EuroMax in Rotterdam. A variant has been proposed by ZPMC. The horizontal or “Side-Loaded” design has been adopted at Pusan Korea and by the developer of a proposed container terminal at Taranto Italy. The latter has been designed by ATS.
Problems with Existing Designs
The Perpendicular design suffers from a plethora of machines and consequently operations are more complex than required. In particular multiple small machines are needed to move containers to and from the stacks and this leads to congestion which in turn adversely affects Ship to Shore Crane [STS] productivity and well as container velocity. The Horizontal design as implemented at Pusan suffers from the same operational problems, only to a lesser degree. That design maintains the conventional method of moving the containers to and from the quay by employing AGVs or other devices that enter the stacks to receive and deliver containers. ZPMC has attempted to ameliorate this issue by providing a complex steel infrastructure between the quay and the CYs. Within that infrastructure, there are numerous rail mounted mini-cranes that act like shuttle carriers to move containers along the quay and effect transfers to automated electric shuttle cars that in turn enter the stacks to be serviced by larger RMGs In all cases the machines employed are r e l a t i v e l y l i g h t weight and carry a correspondingly high maintenance burden. Consequently the Operating Expense [OPEX] of these terminals is not much lower than that of a conventional terminal employing Rubber Tyred Gantry cranes [RTGs]. Additionally, the ZPMC design carries a significant additional Capital Expense [CAPEX] associated with the steel infrastructure as well as the engineering of the quay and adjoining land to carry the dynamic and static loads imposed by the suspended container transfer system.
The Integrated Automated Container Terminal System [IACT]
The IACT is a fully automated method for handling containers in Marine Terminals and Intermodal Facilities that addresses the issues described above. It is comprised of tailored equipment designed to Steel Mill Severe Duty Standards [AISE or CMAA Class F], optimised operating methods and proprietary supervisory software that controls machines and manages the facilities. The system is a 3rd generation implementation based on work by Morgan Crane for Matson Shipping and US Steel. All of the equipment used in the IACT and the Integrated Automated Intermodal Facility [IAIF] is well proven in heavy industry, particularly in steel mill hot metal applications. The configuration of the particular machines and cranes is dictated by operating requirements. A ‘Leap Ahead’ is provided by the Advanced Autonomous Terminal Operating System [AATOS]. A System Supervisory Controller [SSC] employs a variant of the Genetic Algorithm and seamlessly integrates machine management with planning, operating, and management functions into a single system that combines the automatic management functions with machine control. The result is a Real-Time Process Control method that continuously optimises the facility through an iterative process that balances schedules, container inventory, machine functions, maintenance and manning. The IACT is a fully integrated mechanical/software system that operates autonomously to receive, store and deliver containers to and from ships, trucks and trains. The principal mechanical elements of the IACT are shown in Figure 01. Prominent in the design are large [100 meter span] RMGs and associated conveyors. By adopting large robust equipment, we simultaneously reduce the capital and operating expense by reducing the number of machines employed in the terminal.
INSERT Figure 01.
A typical import sequence proceeds from left to right of the drawing and is:
• Containers are picked from the ship by the Ship-to-Shore Crane [STS] and deposited on the Quay Conveyor at the most seaward position on the Conveyor. This is an indexed position known to the System Supervisory Controller and communicated to the STS.
[Note: all pick/deck positions in the IACT Facility are indexed and known to the System Supervisory Controller]
• The Quay Conveyor transports the containers toward the first CY – which is called the’Waterside Yard’
• The RMG picks the container from the Quay Conveyor landside end and deposits it in the Waterside Yard
• When directed by the Supervisory Controller, the waterside RMG moves the container to the Yard Conveyor
• The Yard Conveyor transports the container toward the Land-side Yard;
• The RMG in the Land-side Yard picks the container from the Conveyor and places it in the Land-side CY
• At the appropriate time, the container is moved to the Truck Load/Unload Station for delivery to the Trucker
• The Trucker is directed to the appropriate Load/Unload Station by the Supervisory
Controller and the container is delivered to the Trucker
Two other features should be noted; a Shuttle Car System between the Quay Conveyor and the Waterside RMG allows shifting of containers between hatches of the same ship or between berths for trans-shipment operations. The Shuttle Cars can also be used to quickly reconfigure the CYs in the event that berth assignments change. This IACT arrangement also includes true On-Dock Rail. That is, rail cars can be loaded and discharged on the Berth with the same container handling equipment without an intervening drayage move by Chassis Hustlers or Trucks. Because On-Dock Rail operations are intermittent, containers can be stored in the CY over the rail tracks when no trains are in the facility.
Maintenance burden reduction
All machines are designed and built to Severe Duty Steel Mill Standards of the AISE or CMAA [Class F]. These standards require that the machines run continuously at or near design capacity for a minimum of 25 years. The result is a robust, highly reliable system with extremely low maintenance requirements. Reliability is enhanced by providing numerous work-arounds and designed-in excess capacity. By way of comparison a container terminal employing RTGs and operating 4,500 hours per annum will accumulate about 1 hour of maintenance per shift per day on the RTGs. The maintenance burden for the RMGs employed in the Perpendicular designs described above is slightly lower. For comparison, at the US Steel facility, the RMGs accumulate approximately 1 hour of maintenance per shift per month. That facility has been operating for more than 20 years and has never had a forced outage, a damaged container or a damaged spreader. No container terminal can make
similar claims.
IACT Operating System
Real-time control and management of an automated system such as IACT is beyond the capability of human operators. It requires a fully automated planning computer system that communicates with all equipment and has access to the backend of the customer’s terminal planning and management database.
Systems Supervisory Controller [SSC]
Automated Container Terminal Inc (ATS) has employed the latest thinking in Operations Research combined with intimate knowledge of Container Terminal Operations to develop an advanced autonomous Terminal Operating System, a component of which is the SSC. The SSC employs a variant of the Genetic Algorithm and seamlessly integrates machine management with planning, operating, and management functions into a single system that integrates the automatic management functions with machine control. The result is a Real-Time Process Control method that continuously optimises the facility through an iterative process that balances schedules, container inventory, machine functions, maintenance and manning. The System employs a long-horizon, short-cycle optimisation algorithm that creates work orders for all machines and ensures every container arrives at its designated location at the right time and in the right sequence. A long horizon means that the computer prepares an integrated plan for several hours or even days into the future, depending on the available data, in order to estimate and optimise the future effects of the current plan. A short cycle means that the s y s t e m m o n i t o r s t h e execution of the current plan, by comparing the actual execution times to the work orders and by monitoring for new events that affect the plan. As long as there are no deviations, the system continues the current plan. On the other hand, the system may create a new plan at any time, depending on the need, in order to prevent delays caused by poor synchronisation or e x c e p t i o n s s u c h a s unexpected arrival of trucks or containers. Whenever a new plan is needed, the planning module retains a consistent set of work orders including those being executed, those to start shortly, and those needed to synchronise with them. It simulates the completion of the retained work orders to determine the states (times and p o s i t i o n s ) f o r a l l machines and containers once their retained work orders are complete. Given these predicted states, the planning module repeats the long-horizon plan computation, again optimising over a long period horizon. We create the long-horizon plan using a genetic algorithm. Genetic algorithms are named for the mutation and selection process that occurs in nature. The genetic algorithm manipulates a key vector, which is simply a list of parameters that guide the development of a single plan by a largely heuristic algorithm. Using a heuristic algorithm does not require that the process be modeled in any specific mathematical representation, as it would be using, for instance, integer programming. Heuristics allow the system to take careful account of even the smallest details of machine behavior, yard layout, and other factors, in a straightforward and understandable way. Conversely, the genetic algorithm needs to know nothing about the construction or behavior of the heuristic planner. It only manipulates the key vector and observes the evaluated result of each plan. This evaluated result is a single “penalty” value that the genetic algorithm wants to reduce to the smallest possible value. We use a weighted combination of penalties such as truck wait time, quay crane wait time, yard shuffles, and crane efficiency. After creating a new plan based on a new key vector, the heuristic planner calculates each of these performance values given its predicted results. In the prototype we allow the Automation System Supervisor [a human being] to modify weights to be used as multipliers for the penalties. So, if the optimiser creates plans that have too much truck wait time for instance, the user can increase that penalty multiplier and reduce truck-wait time at the expense of other penalties. The genetic algorithm creates tens or hundreds of complete plans, searching for the best one. As it proceeds, it always retains the key vector associated with the lowest-penalty solution. One could build an algorithm that would simply create new key vectors completely at random and eventually find a very good plan, but that would not be very efficient. By using the described approach, we are able to create re-plans in a few seconds as opposed to tens of minutes or hours. The genetic algorithm also manipulates the key vectors using a random process, but in a more orderly fashion, by creating new plans in even sets, perhaps 8 at a time. After making the first two sets (16 plans) it selects the lowest-penalty 8, from which it selects 4 pairs that we call “parents.” It then mimics the biological process of gene splitting by swapping some of the elements of the key vector between the two parents in each pair. So, one of the parent pairs may originally have keys ABCDEFG and HIJKLMN, but from them the algorithm creates two “children” with keys ABCDLMN and HIJKEFG. It then “mutates” each element of the child vectors by making very small, pseudorandom changes. In this way the children are very similar to the parents but not identical. These new 8 child vectors create 8 new plans and the process repeats, beginning with the new set of 16, all the time remembering the key vector with the smallest evaluated penalty. To prevent iterations from continuing forever, the algorithm stops making new plans based on a simple set of rules. These rules are needed because, with the genetic algorithm, there is no way to know for sure that the mathematically optimal solution (if there is such a thing here) has been achieved. Because we are “killing off” the lower performing children, we mimic the Darwinian construct of natural selection. In fact genetic algorithms are often considered to be within the class of “Evolutionary Algorithms.”
In the context of Container Terminal design, for purposes of evaluation and demonstration, we create a complete simulation model of a one-berth problem, which then activates the work orders from the optimal plan, in a time-synchronous fashion. We visualise the results using a three dimensional display that can be rotated and zoomed at will during the replay. The simulation can also be paused and restarted, either at the beginning or at selected points during the evolution.
We continue the development of the simulation during the design process, refining the details to allow designers to observe the consequences of design decisions. We also utilise the simulation and visualisation during actual operations, allowing operators to examine the behavior of automatic plans in advance of their execution.
Order of Magnitude Cost Estimate by Equipment Type
Because we replace many small high maintenance machines with a limited number of larger more robust machines, a typical IACT when compared to so-called Perpendicular Designs will exhibit a capital cost approximately 20% less than that of the competing design. In the same facility foot print the IACT will double the throughput and increase capacity by a factor of 2 or more.
This article is an abridged version of a paper given at the Port & Terminal Technology Conference, held in Texas, Houston in April 2011, by Dr Joseph H Discenza, Chief Operating Officer, Automated Container Systems Inc., USA.