Chemical reactors are one of the most flexible assets in a manufacturing plant. Changing reactor conditions, feed blend, recipe, or line-up can result in the same equipment producing vastly different products. The transition period experienced between successful production of different products is a side effect of this flexibility.
Transitioning a reactor from one product to another is a costly operation because some material will be produced that falls outside of the specification windows for either product. This off-specification (off-spec) or wide-specification material is sold at a severely degraded margin. Therefore, it is in a process manufacturer’s best interest to minimise both the time taken to transition and the material produced during transition periods to maximise profitability.
This type of optimisation analysis requires searching through years of data to identify specific types of transitions, overlay numerous reactor and laboratory parameters during those transitions, and calculate key performance indicators (KPIs) to rank the effectiveness of the reactor transition. This type of analysis was not feasible at scale before the availability of advanced analytics software applications.
Data consolidation and alignment
Reactor process data tends to reside in a process data historian, while analytical laboratory data typically resides in a SQL-based laboratory information management system (LIMS). Historically, a subject matter expert (SME) analysing data from both sources would use a spreadsheet, with multiple database add-ins and queries required to obtain a fixed range of data. Once that data was obtained, the SME was either dealing with timestamp mismatch on a raw data pull, or potential misinformation from gridding of data at fixed intervals.
Seeq Cortex enables simultaneous access to a company’s different data sources and data types, and makes this data available to the SME in the search bar of Seeq Workbench. The data is accessible exactly as it is stored in the data source, with no loss of data integrity from gridding or interpolation. Furthermore, the data is automatically aligned in time, meaning that the matching of time stamps between high and low frequency data sources is not required. When a new date range of interest comes up, there is no refresh of data pulls and corresponding figures. Instead, the SME simply adjusts the date range of interest, either manually or via configured auto-updates, and the calculations are automatically applied.
While this may sound trivial, the amount of time it can take for an SME to query a process historian, wait while the data pulls, then return to query the LIMS database, can add up to a large portion of a day. In fact, up to 75% of the time required for analytics is often spent gathering, organising, and cleansing the data. This process can be particularly lengthy for reactor transition analysis because SMEs must often look back through multiple years of data to find two or more instances of a unique transition.
Identification of transition periods of interest
The number of product types produced by a chemical reactor production unit is constantly evolving, with new products being developed to meet customer demand. For a production line that makes n different products, this can mean n* (n - 1) unique product transitions.
Comparing two random transitions is akin to comparing apples and oranges. To make an insightful and actionable comparison between transitions, an SME must identify transitions between the same initial product type and final product type, while weeding out transitions influenced by outside factors such as shutdowns or turnarounds. Some transition types may be common, while some may only occur once every year or two. To avoid hitting the row limit in a spreadsheet, this required first pulling years of low frequency data, filtering and manipulating the data in an attempt to locate some transition times of interest, focusing time ranges for multiple higher frequency data pulls, querying and aligning laboratory data to the process data timestamps, and only then beginning to analyse the data. When a new transition of interest popped up, the whole process was repeated.
Figure 1. Seeq cheat sheets provide SMEs with a quick guide to the different techniques they can use to define conditions.
New data types, such as Seeq’s ‘capsule’, can represent a singular time period of interest. A collection of similarly defined capsules is termed a ‘condition’. These conditions can be defined by a signal exceeding a threshold, an increasing/decreasing trend or step change, the time period when a piece of equipment is running, the night shift hours, or other parameters (Figure 1). Any number of conditions can be combined to define the specific event an SME is trying to capture. An example of a condition built around a specific online transition might look like a composite of the following:
- There is a step change in product type signal or reactor temperature and feed blend set points.
- The initial product type was GRADE123.
- The final product type was GRADE456.
- The feed to the reactor was above 5000 kg/hr for the duration.
- The product specification parameters were not within range for the product type being produced.
The creation of finely tuned conditions enables the next analysis steps.
Leveraging conditions to calculate key metrics
Capsules can store metadata called ‘properties’ that can be used to filter the condition for transitions of interest, or to perform broader aggregations across all transition types. From this early point in the analysis, users can gain valuable insights, such as how often a given product transition occurs, as demonstrated in the histogram shown in Figure 2.
Figure 2. Histogram showing count of various reactor transition types over a six-year time range.
Filtering for a specific transition of interest enables SMEs to extract full value from the ‘Workbench Capsule Time’ feature by overlaying process signals and laboratory data during similar transitions (Figure 3). From this view, SMEs can identify and evaluate the impact of different starting set points, controller ramp rates, and process variable adjustments. This visualisation can also serve as a starting point for the calculation of an ideal transition profile, created from identified best transitions using the ‘Workbench Reference Profile’ tool, such that future transitions of that type can be monitored against the corresponding statistical profile in near real-time.
Some of the most valuable reactor transition key performance indicators (KPIs) require additional tuning of a transition condition to capture the exact duration between when product A went off-spec and product B came on-spec. This is achieved in Seeq using a combination of point and click tools and free-form formulas, providing the connection of points in time between the last on-spec sample of product A and the first on-spec sample of product B.
Figure 3. Process parameters are easily overlaid for equivalent transitions with capsule time.
This finalised transition condition can be used to calculate the average, standard deviation, minimum, and maximum transition duration. This information can be used as a target to improve upon, as well as a benchmark for determining when a transition is approaching an abnormally long duration, requiring further investigation. Production rates can be aggregated to calculate the average throughput rate during the transition, or the total volume of off-spec material produced. The information provided by this relatively simple statistical analysis can serve as a gateway for transition rate optimisation, along with supply chain and production cycle optimisation.
Extended applications of results
The identification of transition capsules and calculation of transition KPIs would likely be executed by SMEs. While there is significant value to be gained from this analysis alone, additional benefit could be unlocked through collaboration with data science and business teams.
Data scientists leverage programming languages to solve optimisation problems. There is an interesting challenge that surfaces when attempting to pick a target production rate for the transition. Lower rates mean lower volumes of off-spec material produced per unit time. Lower rates also mean longer residence times in equipment, and greater time required for the setpoint adjustments made during the transition to be realised by the final, tested product.
Data science teams can make informed hypotheses using machine learning algorithms about the relationships and interdependencies among production rate, residence time, transition duration, and transition volumes. When the data set is presented to them after being analysed and contextualised by an SME to focus on a subset of product transitions, they can begin their analysis and arrive at conclusions sooner.
Different chemical reactor products may require different feed blends, catalysts, packaging, and shipping transportation. This presents difficulties in adopting a just-in-time production process that can rapidly pivot to meet changing customer demand. Instead, a product cycle or wheel is often utilised by supply chain organisations to balance variable customer demand with expensive inventory costs. The analysis of reactor transitions is an invaluable starting point when establishing a new product cycle, evaluating an existing cycle, or calculating the financial impact of a demand-driven cycle break-in.
Supply chain and sales teams can work directly with customers to understand the frequency and volume of product shipments. SMEs can directly apply the results of this analysis when collaborating with supply chain teams to generate a production schedule meeting customer needs, while minimising the amount of off-specification time and material.
Advanced analytics applications such as Seeq are significantly lowering the time and effort required to gain insight into the historically murky waters of the transition period between chemical reaction products, providing value to process manufacturers.
Written by Allison Buenemann, Seeq Corp., USA.
Read the article online at: https://www.hydrocarbonengineering.com/special-reports/13052021/gaining-insight-into-murky-waters/