Case Studies
A power plant uses up some of the electricity it produces for its own operations. In particular the pumps in the cooling system and the fans in the cooling tower use up significant amounts of electricity. It will increase the effective efficiency of the power plant if this internal electricity demand can be reduced.
For the particular power plant in question here, there are six pumps (two pumps each with, respectively, 1100, 200 and 55 kW of power demand) and there are eight fans, each with 200 kW of power demand.
The influence we can exert on this internal electricity usage is to be able to switch on and off any of the pumps and fans as we please – with the restriction that the power plant as a whole must be able to perform its intended function. A further restriction is introduced by allowing a pump to be switched on or off only if it has not been switched in the previous15 minutes. This is to prevent too frequent turning off and on.
A rod pump is a simple device that is used the world over to pump for oil on land, see figures 1 and 2. Basically, we drill a hole into the ground and cement the hole (with a well casing) so that a nice vertical cavity results. Into this cavity a rod is inserted that is going to move up and down using a mechanical device that is called the rod pump. Attached to the bottom of the rod is a plunger that is a cylindrical "bottle" used to transport the oil. On the downward stroke, the plunger is allowed to be filled with oil, and on the upward stroke this oil is transported to the top where it is extracted and put into barrels.
The injection molding technology is a widely used technology for the mass production of components with complex geometries. Almost any material can be processed with this technology. For polymers the pelletized/fragmentedmaterial is injected into a hollow mold cavity (showing the negative structure of the resulting part) and then molded under elevated temperature. The part is cooled down and ejected.
We propose that popular benchmarking studies can be augmented or even replaced by a method we call self-benchmarking. This compares the present state of an industrial plant to a state in the past of the same plant. In this way, the two states are known to be comparable and it is possible to interpret any changes that have taken place. The approach is based on data-mining and thus can be run regularly in an automated fashion. This makes it much faster, cheaper and more meaningful than regular benchmarking. We demonstrate by the example of maintenance in the chemical industry that this approach can yield very useful and practical results for the plant.
A power plant’s turbine is its most important component. A turbine failure can lead to a long-term shutdown of the entire plant. Thus, it is important to carefully watch the turbine for any signs of abnormal behavior using a variety of sensors installed in key locations on the turbine. The most crucial information regarding the health of the turbine is contained in the vibration measurements. All sensor output is logged into a database and therefore available for study.In order to prevent an expensive shutdown through an unexpected turbine failure, it is expedient to monitor any abnormal behavior that may occur and that could be extracted from the recorded data.
In this study we will demonstrate that it is possible to predict a known turbine failure using historical data. On a particular turbine, a blade tore off and completely damaged the turbine, requiring extensive and expensive repair and replacement. After the event, the question was raised whether this failure could have been predicted and localized to a specific place inside the turbine.
The specific turbine in question has over 80 measurements on it that were considered worthwhile to monitor. Most of these were vibrations, but there were also some temperatures, pressures and electrical values. A history of six months was deemed long enough, and the frequency of measurement depended upon each individual measurement point – some were measured several times per second, others only once every few hours. In fact, the data historian stores a new value in its database only if the new value differs from the last stored value by a predefined parameter. In this way, the history matrix contained a realistic picture of an actual turbine instrumented with sensors as it is normally done in the industry. No enhancements were made to the turbine, its instrumentation or the data itself.
A chemical plant has a particular unit that is meant to combine several chemicals from a variety of input sources in order to provide a gaseous output (called “tailgas”) with a composition that is as constant as possible.In our particular case, this task is run by an assembly of 40 valves that are controlled by a computer that opens and closes them according to a well-balanced schedule. If the valves do not open and close according to schedule, or if they are either too fast or slow, or if they leak, then the tailgas composition is not constant and causes problems later on in the process. In this study, we demonstrate how to predict future problems and to identify the valves responsible for recurring problems in the composition of the tailgas.
A chemical plant’s efficiency and profitability can be optimized using mathematical modeling. The optimization tells plant operators which set-points should be changed to obtain the maximum profitability.This method requires no engineering changes to be made to the plant. We show that a profitability increase of approximately 6% was possible in a specific chemical plant producing silanes, with an overall yield increase of 5.1% and an increase of 2.9% for the most profitable end product.
In a catalytic reactor, at least two substances are brought into contact with each other. One is a substance that we would like to change in some chemical way; and the other substance is the catalyst that is supposed to bring this change about. The two substances are mixed and heated to provide the energy for the change. It is also necessary to provide the plumbing for the substances to enter the reactor and for the end product to leave the reactor. Some parts that are not converted have to be re-cycled back for a second round (and possibly third and more rounds) through the reactor until all of the original substance has finally been transformed. One example for this process is the breaking down of the long molecular chains of crude oil in the effort to make gasoline.
As already indicated, the catalyst performs its work upon the substance and brings about a change. However, by doing so, itages over time and thereby exhausts its potential to cause thechange. This degradation of the catalyst is the primary problem inoperating such a reactor continuously over the long term. The catalyst must therefore be re-activated in some fashion and at some time.
The efficiency of the combined-heat-and-power coal-fired power plant Reuter-West in Berlin, Germany can be increased by 1.1% using mathematical modeling of its processes.
The Vattenfall power plant Reuter-West in Berlin, Germany, has an efficiency that depends on how the plant’s different processes are run. While many smaller processes are automated using various technologies, the overall process is largely controlled by human operators. Therefore, the maximum possible efficiency of the plant depends partiallyon the decisions, knowledge and experience of the operators. There are two main challenges for them:
In the maintenance department of a large chemical facility with several plants, we are faced with the problem of budget planning for the future. In this particular case, we are analyzing the last ten years. The first five years were spent doing the maintenance in-house and the next five were spent with the maintenance outsourced to a service provider. This service provider has a fixed-fee contract so that a single yearly price pays for all necessary maintenance activities required so that the plant is running at a certain availability (or better). We therefore have all relevant data, both technical and economic, only for the first five years. For the second five years, we have only the technical data.
At a major European wholesale retailer, hoteliers, restaurants, caterers, canteens, small- and medium size retailers as well as service companies and businesses of all kinds find everything they need to run their daily business. Every customer has a membership number and card. Due to this, it is possible to attribute every item sold to a particular customer.
Customer segmentation in general is the problem of grouping a set of customers into meaningful groups based, for instance, on their profession or on their buying behavior. In this particular case, it also allows us to trace which customers belong to which group, because we are aware of their (business) identities. This kind of trace possibility is attempted by many other retailers via loyalty programs in which clients also allow the retailer to attach their identity to the products purchased.
Sometimes offshore oil wells break down and need repair such as when a pump fails. If spare parts are not readily available, this can cause unwanted downtime that is expensive in terms of yield failure and repair costs.We show how costs can be minimized by predicting the status of pumps up to four weeks in advance– allowing preventive maintenance to be performed. This is made possible by using a mathematical model of the pumping operation using automated machine learning methods. This method was applied to shallow-water offshore oil wells in the Dagang oilfield covering 34,629 km2in China. We consider data for 5 oil-wells of a shallow water oil-rig in Dagang operated by PetroChina.
A production plant makes automotive parts on a production line involving many stations. Each station performs one step in the production process. At various stages along the line, we have checking equipment that perform a variety of functional tests on each part. When a part is found to be defective, it is flagged as such and no longer treated on the stations further down the line. If a part makes it to the end of the line without a flag, it is, by definition, a good part because it has passed all testing stations.
If a part is found to be not okay, we know – thanks to the flag – the first reason encountered for it be so. Clearly no process is perfect and so we must expect some scrap parts to be produced. Of course, we would like these cases to stay at a minimum, and so we would like to be able to respond quickly if and when the production line is – for whatever reason – suddenly producing more defective parts than would normally be the case. As the data comes in from production, we would like to know, therefore, if the likelihood of producing “flagged parts” has recently increased– for whatever reason – or has not increased.
Wind power plants sometimes shut down due to diverse failure mechanisms and must therefore be maintained. These maintenance activities are costly due to logistics and delay–especially in the offshore sector, but also in the countryside. Common failures are, for example, due to insufficient lubrication or bearing damages. These can be seen in vibration patterns if the signal is analyzed appropriately.
It is possible to model dynamic evolving mechanisms of aging in a mathematical form so that a reliable prediction of a future failure can be computed. For example, we can say that a bearing will fail within 59 hours from now because the vibration will then exceed the allowed limits. This information allows a maintenance activity to be planned in advance and thus saves collateral damage and a longer outage.
In a nuclear power plant, if the vibration of the turbine axle exceeds a certain limit, we speak of a vibration crisis.It does not represent a real damage but it could, if left unchecked, lead to a major failure. The exact cause of the problem is not precisely identified at present but it always occurs during the same conditions of vacuum pressure and power, two essential measurements on the plant.
This study concerns itself with the prediction of future vibration crises and not with determining the mechanism that causes such vibrations. If one could know hours in advance that a crisis will happen, this would help operators to alleviate it: The plant can be regulated into a state more conducive to controlling the impending crisis.
Condition Monitoring usually analyzes each measurement separately using static limit information. This results in false alarms and unhealthy conditions that are not alarmed. Using machine learning techniques, the big data gathered around large equipment or an entire plant can be analyzed as a single coherent whole to draw conclusions about its current state of health. First, a mathematical model of the relevant measurement is created using the other measurements available. This model represents the equipment or plant as a unit when it is operating as it should. Second, this expected value is compared to the measured value. If they agree, the current state is healthy. If they do not, an alarm is released and a maintenance activity must follow. This method is seen to be far more successful than standard condition monitoring thus preventing false alarms and always alarming unhealthy states.
