BULKtalk, Engineering

BULKtalk: Availability, reliability, utilisation and throughput

Steve Davis, Senior Bulk Handling Expert at Advisian was recently the subject matter expert on a study for a terminal upgrade. He shares some of his findings on how to improve plant design to increase availability and throughput.

As the subject matter expert on a study for a terminal upgrade, one of the key requests we received was to confirm the overall plant availability and throughput. After a few years of operation, the owner was struggling to handle 20 per cent of the perceived nameplate throughput.

This leads to some interesting observations. The first, and perhaps most misunderstood aspect, is that the engineering, procurement and construction (EPC) contractors had agreed to the specified facility availability of 98 per cent. The owner assumed this meant that the plant has a capacity of 98 per cent of nameplate (operating hours multiplied by nameplate tonnes per hour).

This shows a lack of understanding of availability by both parties, as 98 per cent is unrealistic for a single stream operation with a railcar dumper, multiple conveyors, storage with stacker/reclaimers, more conveyors and ship loading all arranged in a series with no redundancy.

Other misunderstandings included definition of availability in the plant as utilised, what reliability impacts should be considered and mitigated, and understanding that attaining the required throughput is key to a successful operation.

Availability and utilisation

So, what is availability? It is the percentage of time that the plant remains operational under normal circumstances in order to serve its intended purpose.

For an approximation of availability of a linear system we can multiply the individual components’ availabilities. If there are 10 conveyors in a series and availability is assumed at 99.5 per cent, or 0.995, a car dumper at 0.95 and a stacker at 0.9, we get an overall availability of 0.99510 x 0.95 x 0.9 = 0.813, or 81.3 per cent availability for the system. To obtain the required 98 per cent system availability each component must be 99.85 per cent available, i.e. 13 hours per year unavailable on average. (We assume individual equipment components and failures over time result in these availabilities, this can be calculated if accurate maintenance data is available). Redundant systems can also be approximated.

The formula is given by: 

Here we had 80 per cent availability given by

For this plant, could we expect to operate for 80 per cent of the time and therefore get 80 per cent of the nameplate throughput? Here are some thoughts:

The plant is a batch operation, and as such only operates for approximately 50 per cent of the time when trains arrive to discharge, and when ships arrive to load. Some repairs and maintenance that would have added to the downtime are completed in the unutilised periods.

Utilisation is therefore 50 per cent, so is potential throughput now only 40 per cent of the nameplate – 80 per cent availability x 50 per cent utilisation?

Train unloading is a continuous operation and manages the design rate.

The shipload average/aggregate loading rate is always less than the design ship loading rate, due to hatch changes and other delays, unless a surge bin system and surge rated ship loader are included. The plant does not have these surge systems and the shipload rate can only average approximately 75 per cent of nameplate, and therefore is utilised for 33 per cent more time than expected by the owner. This does not limit plant throughput as the ship loader in this study has spare capacity at the current throughput but does reduce the expected expansion capability.

The imbalance between train delivery rate and ship loading rate is handled by the stockpile.


Occupancy for batch systems such as train dumping and ship loading are aspects that must be considered.

A car dumper or ship loader is occupied for the entire period when no other train or ship can use the facility:

  • Plant design considered four trains per day, with three trainsets. The design assumed train overall cycle time of 12 hours, which allows six hours per day per train for fuel, maintenance checks and driver changes. Owner imposed checks on loading and unloading extended the cycle to 18 hours, reducing the number of potential trips by almost one each day or reducing maintenance time. This was sufficient to unload the demand, but all spare capacity has been used.
    A fourth train and second car dumper is being considered to provide catch up capacity and redundancy.
  • Ship loading in this project has many occupancy facets contributing to lower than design occupancy.
  • Shipping constraints require 15 hours between a ship departing loaded and the next ship arriving to load. Design allowed for six hours.
  • Average ship sizes are 60 per cent of the design size, therefore there are 67 per cent more ships loaded each year than design. The design assumed 80 ships and the actual is 133. The berth is occupied for much longer than considered in design.
  • Many of the smaller ships cannot deballast at a rate matching design load rate, so the length of time loading is extended. On average by 50 per cent of design.

Ship loader occupancy was calculated on a corrected basis, 91 per cent. This results in 5.8-hour average between ships, and does not give time for cleaning, breakdowns, maintenance, upsets in the shipping schedule, and the like. One significant failure effectively brought the plant to a standstill as ships could not be loaded.

The owner is assessing the available fleet to reduce the number of ships and assessing options to reduce the 15 hour delay between ships and expects to reduce occupancy to 75 per cent or less. Increasing the ship loader system design rate will be complex and expensive. An additional ship loader is being considered to provide additional capacity and redundancy.

The following issues are identified:

  • Little understanding of availability, utilisation, capacity, batch operation of a materials handling system by EPC or owner resulting in significant difference in capacity versus expectations.
  • Many operational impacts on the operational aspects by owner without considering these factors in the design.
  • A good materials handling design should have consulted with the owner to identify these aspects and the impacts during feasibility and provide a better solution.
  • The real capacity of rail unloading has no spare capacity.
  • Shiploading at 91 percent occupancy is untenable. The target was 50 percent.

Catch up capacity

A further issue that should have been considered in detail in this system was the need for catch-up capacity. The rail system as designed effectively has a spare train at the design throughput. This means that should there be an upset and it is not possible to transport to the terminal, product can be stored temporarily at the loading point. When the system is running again the trains run at full capacity and the load point storage is drawn down at a faster rate. In this case three trains could transport 133 per cent (33 percent catch up) of steady demand as designed, however the spare capacity has been taken by the longer cycle time.

Similar considerations are relevant with the ship loader. The original design considered under 50 per cent occupancy and significant catch-up capacity. Operating at 91 per cent occupancy makes it difficult to meet the nameplate and there is no catch up capacity.

Downtime and outage records, availability and reliability

The basis for building reliability, availability and maintainability (RAM) models is the data that is recorded or developed as input. This plant had three years of data, which is insufficient as an indicator or to develop model input data. The data did indicate types and durations of the outages, but there are doubts on the accuracy and completeness of the data. Our model was therefore based on a combination of available data and previous models for similar equipment.

Recorded downtime was mostly short unplanned failures of idlers, bearings and other items, spurious trips and the like. Downtime was recorded when the plant was being utilised. The 1750 hours downtime each year consisted of just over 900 outages, average just under two hours each, and roughly 2.5 every day. Fewer than 10 outages were longer, reaching a maximum of four days. The downtime data does not consider the loss in throughput every time the system ramps down and back up, and the potential for the system to fail to restart after repair.

The downtime recording methodology was ineffective and made it impossible to extract other than average data. Poor data are the result and poor training and implementation. It seemed that every input was by a different person with a different perspective on the cause, description and the time to repair.

For example, idlers are a common point of failure. We extracted more than 700 idler failures in the three-year data pack. Idlers were only identified as to conveyor and not to type or location, so repeat failures could not be identified for improvement. In the 700 failures, we had more than 250 separate descriptions of the failure, none of which was useful in identifying any root cause, and which makes data interrogation difficult. Statements such as ‘idler failed’, ‘idler replaced’, ‘broken idler’, ‘defective idler’ with many variations are useless for assessment. The variation in description makes it difficult to extract similar failures, and instead of sorting spreadsheet data, manual categorisation is required. Time recorded varied between stop to restart and repair time only – many times appeared to be based on complete shifts rather than actual time. From discussion there were many failures that had not been recorded.

The recording system had been set up with 28 items of data to be entered for each outage and little guidance on what to enter. I believe meaningful data can be collected in less than 10 items. Each cell for data should have a pull-down menu that limits what can be entered.

Reliability is often measured as mean time between failure (MTBF) and is the number of failures in the total time available. This is often recorded per equipment item, and not for the entire plant. In this case, each of the 31 machines (dumper, conveyors, stackers etc.) in sequence was relatively similar and accounted for about 30 outages a year in each machine. It is difficult to assess actual MTBF if the failures and times are not accurately recorded.

The MTBF was 8760÷30 = 290 hours per machine. Or should it be half this, as utilisation is 50 per cent? One hundred and forty-five hours indicates one failure in each machine every six days on average.

If we consider the whole plant system operating at 50 per cent to 75 per cent utilisation (when the failures occur), MTBF is between 8760 x 0.5÷900 = 5 hours and 8,760 x 0.75 ÷ 900 = 7.3 hours or one failure on average every five to eight operating hours. This plant is not very reliable.

Low reliability in this case was the result of:

  • Higher than design occupancy of equipment leaving less time for maintenance
  • Selection of inappropriate equipment and components for environment and duty
  • No condition monitoring strategy
  • Lack of operator training in operation and maintenance
  • Owner and EPC unfamiliar with design of a batch bulk handling system


Availability and reliability are real indicators of plant health. However, the key requirement for a bulk handling system is to be able to meet the throughput under normal and upset conditions over a sensible time scale, without bringing any part of the system to a standstill.

We use discrete event simulation software to provide a simulated real-time analysis of the operation of the system. The model incorporates all key components of the system from start to finish. In this case from the manufacturing plant through storage, train loading, rail system, unloading, conveyors to storage, storage, conveyors to ship and the shipping stream.

The model is built with data on operation and maintenance according to design, and then various deviations and upsets are added to see the impact. The model runs a Monte-Carlo iteration over 30 years. In our system, the design met requirements and throughput provided everything was according to the design parameters and reliability was ‘normal’. As soon as the differences to the design basis were introduced the throughput started to drop and various problems are apparent.

If a simulation model had been used before design was completed and a full evaluation of proposed operations had been introduced, many of the system design details would have been different. The facility would have had a simpler arrangement with fewer and different pieces of equipment, and it would have easily handled the expected throughput.

Send this to a friend