As the current manufacturing environment exhibits market and price competitiveness, there is an ever-increasing need to produce quality products at a lower cost to meet the market demands. While addressing these demands, manufacturing plants are faced with significant challenges. One of these challenges is the increased cost of production resulting from high maintenance costs due to frequent and costly failures of equipment. However, throughout the cycle of production and maintenance, manufacturing plants generate and collect large amounts of data that could be leveraged to motivate decisions that add value to their maintenance procedures and operations.
The subject of this research is a plant that is faced with similar equipment availability challenges, yet generates and collects several different types of data. However, the plant is unable to use all the different data sources to motivate comprehensive maintenance decision.
4.1 Plant Background
The demand for the analysis stems from a Portland cement production facility located in Kenya, which for the past three years, has built a collection of equipment stoppage records and vibration readings for several sections of the plant. Portland cement, which is the basis of concrete, is produced through a cyclic closely-controlled process of crushing, mixing, and heating combinations of mined materials. As the production process is both resource intensive and dependent upon material availability, each stage of the production cycle is particularly vulnerable to stoppages. The company under study operates a cement production plant that is organized into 8 sections, each corresponding to a different phase of the production process. Each section is further comprised of physical equipment, each uniquely identifiable.
A high-level overview of the cement plant layout is shown in Figure 4.1, which identifies the 7 plant sections(Section 1, where raw materials are mined from a local quarry, is not part of the manufacturing process and is not included in the research) involved in the cement manufacturing process. The plant is organized into these sections according to the respective manufacturing function, as identified in Table 4.1, with each section containing equipment specialized for the specific function.
|02||Raw material preparation||06||Cement grinding|
|03||Raw meal grinding||07||Cement dispatch|
|04||Raw meal homogenizing||08||General plant services|
4.2 Problem environment and statement
In its current operational state, the cement plant faces several challenges which this research will aim to address. The plant keeps records of all equipment stoppage events(both failure and non-failure events) and equipment vibration readings, in addition to tracking monthly production totals. Despite historical records, the data is only presented in weekly or monthly reports for the purpose of calculating overall plant availability. As the stoppage event records are segmented into weekly reports, the historical information is distributed between a large number a files. This structure prevents the stoppage data from being used for analysis to facilitate maintenance decision support.
Despite the use of corrective, condition-based, and preventive maintenance, the plant continues to experience frequent failures of the equipment. Additionally, the plant experiences a considerable number of non-failure related stoppages, such as lack of raw materials, lack of power, fuel, and other process related consumables, which may adversely effect the maintenance and maintainability of the equipment. Unfortunately, the organization lacks a framework to establish how the different stoppage events(failure and non-failure) impact the maintenance of the plant.
Since each data source is generated from a different process, the data is collected independently and at different levels of abstraction, and is analyzed independently using different methods. For example, stoppage events are tracked to a specific equipment, but the broken part or component is not uniquely identifiable. Additionally, production output is tracked at the plant section or sub-section level, but not the equipment level. One equipment may be monitored for vibrations on several different parts, and another equipment on a different set of parts. As the different data sources may be relevant to different levels of the plant, it prevents the data from being readily used when necessary.
Although historical records pertaining to individual equipment items are maintained, the organization only derives availability measured at the overall plant level. Stoppage events are not analyzed to identify critical sections of equipment or to model long-term reliability characteristics. Despite maintaining the historical records, the plant has no decision support framework for using this information to guide maintenance decisions.
Additionally, the historical production data and vibration readings are stored in structures that neither facilitate inference nor integration with other sources. As the data is largely unstructured, the plant is unable to integrate and utilize all the data sets, and requires extensive work to devise and implement a cohesive structure. However, once integrated, it will be possible to perform knowledge extraction on the data while in a combined state. After integrating the data sources and extracting all available information, it can be used to develop predictive models to derive maintenance decision support.
In acknowledgement of these challenges, the organization has a desire to improve and optimize their maintenance programs through the use of data-driven decision support. Based on these challenges, we derive our research objectives as described in the next section.
The primary objective of this thesis is to develop an integrated predictive model, incorporating failure event records, production output records, and vibration observations, that can be used to predict the reliability and behavior of the mechanical equipment of the plant. In order to achieve the primary objective, the research must accomplish several specific objectives.
The first specific objective is to undertake data pre-processing to prepare raw data for the current analysis, as well as future analysis. This will involve aggregating the data from each source into a single repository, cleaning the repository and removing non-informative formatting, transforming the repositories into functional tables, and standardizing the data according to relevant standards. This objective includes repeated consultations with domain experts at the cement facility to obtain clarifications regarding data structures and relevant terminology.
The second specific objective is to perform a descriptive analysis to identify important characteristics regarding the scope of the data collected. Additionally, a criticality analysis will be performed to identify key sections and equipment items within the plant on which to focus for subsequent analysis.
After identifying several critical elements of the plant, the third specific objective is to build reliability models that can be used by a maintenance engineer for maintenance decision support. Such a model would identify specific reliability characteristics of a given equipment, demonstrating a methodology that can be applied to any area of the plant in future work.
The final specific objective involves an integration of the failure event records, the production output records, and the condition monitoring(vibration measurements) records into a predictive model. The integrated data will represent the maximally available knowledge regarding the condition and behavior of an equipment with which to predict future events, and prescribe future maintenance.
As mentioned previously, the cement plant maintains records detailing all stoppage events occurring within the plant, a record of monthly production totals according to plant sub-section, and a record of vibration readings measured via a handheld probe.
The equipment stoppages are recorded on a daily basis (for weekly reporting), with each stoppage linked to a uniquely identifiable equipment code. The stop time, start time, and total duration are recorded for each stoppage event, along with a brief root cause description. Additionally, each stoppage is further categorized by a code indicating the type of stoppage that occurred(e.g., planned maintenance, mechanical failure, etc.). Furthermore, each entry contains a free-form field in which the maintenance engineer can include a comment explaining what may have occurred and what action was taken to resolve the stoppage. As the stoppage events have been recorded over a three year period(2015, 2016, and 2017) the plant has experienced a total of 30,380 stoppage events.
In addition, production figures are recorded in the form of total number of tons of material processed per month. For sections(e.g., cement grinding) which are made of up multiple sub-section running in parallel, the production figures are recorded for each sub-section. For comparison, the plant has also provided the maximum production capacity rate(in tons per hour) for each of the respective sub-sections.
Finally, the plant also maintains a record of vibration measurements which are taken directly at the equipment level with a handheld proprietary probe. Vibrations are not monitored continuously, and are done so at irregular intervals, as deemed important by the plant, and not all equipment items are monitored. In total, there are 1,634 available vibration observations taken throughout the same three year period.
4.5 Research direction and structure
This thesis will detail the research of a multidimensional reliability analysis of mechanical equipment being used in a cement production facility. The methodology and subsequent results will focus on providing decision support for maintenance actions performed on the equipment, and to characterize the effect of these actions on reliability. This paper will detail each step of the research process, from cleaning and transforming the observed data, to comparing statistical models, and interpreting results.
Following the introduction of the objectives and scope of the research, the second chapter will provide a summary of academic works and techniques that may provide background information, justification, or context for the following research. Next, a chapter describing the methodology of the research will detail all of the steps to be taken during the process. After applying the methodology, the resulting analysis will be described, with results and insights explained as they are uncovered by the research. Following the results chapter, a conclusion chapter will summarize the results of the analysis with respect to the research objectives, providing a cohesive assessment of the knowledge that has been gained. Finally, a post-analysis discussion will provide a critical assessment, from the perspective of the researcher, regarding points of improvement for the methodology or research subject, in addition to recommendations for future work.