Food Industry Perspectives on the Implementation of a PAT Strategy

Abstract

This chapter focuses on the drivers and hurdles to the implementation of a process analytical technology (PAT) strategy in the food industry. Case studies, from the dairy processing industry, highlighting the benefits and challenges of adopting a PAT processing philosophy are presented. The case studies demonstrate the potential for improved product consistency and enhanced process control through minimising the variability of critical quality attributes. Advantages of implementing a PAT system including improved product documentation and traceability are also outlined.

12.1 Introduction

The process analytical technology (PAT) tools and analysers have been known within the food industry for decades. Previously, the focus with these tools was to implement on-/in-line analysers in the production process just to have real-time measurements to monitor the production. Within the past couple of years, the focus has changed from implementation of on-/in-line analysers just for monitoring the product attributes to the use of PAT technologies to understand and control the whole manufacturing process and to consistently ensure a predefined quality at the end of the manufacturing process.

Process variation within a manufacturing process is caused by, for example, uncontrolled disturbances like changes in the milk used in cheese making and the temperature of the surroundings. In an ideal world, there would be no changes to the uncontrolled variables. A given combination of settings for the process set points will always result in a product with a given quality. In the real world, the uncontrolled variables will change and a given combination of settings for the process set points will always result in a product that varies in quality. Variation in the product quality can be reduced if it is possible to determine the optimum settings for the process set point during the manufacturing process.

Table 12.1 Examples of PAT tools employed for various operations within the dairy processing industry

Prediction of the optimum settings for the process set points requires many years of experience for the production personnel or a control system that can take the changes to the uncontrolled disturbances into account and predict the set points. Such a control system can be developed based on the PAT tools such as on-/in-line analysers and multivariate data analysis. A schematic view of the principle behind a control system is shown in Fig. 12.1. Here, a manufacturing process with two process steps is shown. An advanced control system can be built for the total process or for each process step. The aims of an advanced control system are to be able to predict the quality of the end product based on knowledge of the raw material and process settings used and to be able to predict the optimum settings for the process set points based on knowledge of the raw material used and specification of the predefined quality of the end product. An advanced control system is based on process models, which relates measurements of raw material and process variables to the outcome of a process or process step. The process model for process step 1 relates measurement of the raw material and the different process variables, like temperature, pH, speed and amounts of materials used in process step 1 with measurements of the semimanufactured product from step 1. The process model can be based on multivariate data analysis , first principle models or a combination. After model validation , this process model can be used to predict the optimum process settings for process step 1 based on measurements of the given raw material used and a given setting for the expected quality of the semimanufactured product. A total process model or a combination of individual process models for steps in a production process could be used to optimize a production process.

Fig. 12.1

Schematic view of the principle behind a control system

An advanced control system could be an ultimate solution for understanding and controlling a manufacturing process. Before it is possible to build a control system fully or partly based on multivariate data analysis , the necessary analysers used to measure the critical quality and control attributes need to be implemented and a system for automatic data collection installed. A PAT strategy needs to be developed and implemented.

This chapter focuses on advantages and challenges when implementing PAT and is based on experience gained during 7 years of work within the area. Case studies from dairy processing will demonstrate the potential for improved product consistency and enhanced process control through minimising the variability of critical quality attributes .

12.2 A PAT Strategy

The drivers for an implementation of a PAT strategy are usually to gain a uniform and high product quality, improve yield and reduce production costs. The propagation of PAT within the manufacturing processes has led to focus on process understanding and continuous improvements. This is due to the fact that it has become possible to generate relevant and high-frequency data by use of, for example, spectroscopic analysers, where a measurement is performed within seconds or minutes instead of using reference methods that take hours. The driver for implementation of PAT must be that the new strategy gives value to the business. An economical driver will result in commitment and focus in an organisation.

Within a manufacturing company, it should be discussed if and how a PAT strategy and the implementation of, for example, on-/in-line analysers gives value to the business. The number of possible PAT cases within a production site will depend on the size and complexity of the production.

During the development of a PAT strategy , one of the focus areas should be to identify possible changes to working routines within the organisation. Implementation of a PAT strategy can affect the routines in production and the laboratory. If changes are identified, the management should handle this and make sure that the routines are changed properly. This is an important task to ensure a successful implementation of a PAT strategy and the future work.

The purpose of a given PAT case has to be described in detail. What is the goal? What are the conditions for a successful application? What are the risks? How can the goal be reached—is it by implementation of an analyser, an advanced control system based on multivariate data analysis or both? What are the milestones? What is the business value?

12.2.1 Business Value Calculations

In the calculations of the business value, all the investments and costs needed and the expected economic gains for the implementation of a PAT strategy are listed and estimated .

• Investments and other costs: e.g. analysers, changes to the production equipment and IT systems, costs of samples taken for calibration of an analyser, time used by production technicians, laboratory technicians, engineers and others during implementation, maintenance costs and also costs related to education of personnel and consultant assistance from suppliers.

• Economic gains: e.g. improved yield due to reduced process variation, increased product quality, reduced amount of product outside specifications, reduced product costs and improved process knowledge.

Estimation of the economic gains is based on analysis of historic process and product data. The historic data are analysed to identify the present process and product variation and from that the potential gain is estimated. Estimation of the value of improved process knowledge can be a challenge, but in spite of that, it is an important gain, which needs to be addressed .

Another economic gain not mentioned above is reduced costs in the laboratory since there will be fewer samples to be measured by the reference method after implementation of an at-/on-/in-line analyser. The reason why it is not included in the list is that the calibration, validation and maintenance of an analyser require resources. Even though there will be fewer samples to measure, other tasks need to be taken care of instead, e.g. recalibration of an analyser.

There are different ways to estimate the business value, but a common estimate to use is the net present value (NPV). From the estimated investments, costs and economic gains, an NPV is calculated. The NPV indicates how much value a given project gives to the manufacturing company within a given period (e.g. 5 years).

The NPV is calculated as (12.1)

$$\text{NPV}={{C}_{0}}+\frac{{{C}_{1}}}{1+{{r}_{1}}}+\frac{{{C}_{2}}}{{{\left( 1+{{r}_{2}} \right)}^{2}}}+\cdots +\frac{{{C}_{T}}}{{{\left( 1+{{r}_{T}} \right)}^{T}}}$$

(12.1)

C ₀ is the cash flow at date 0. C ₀ includes investments and costs of initiating a project and the value will therefore be negative. C ₁, C ₂,…, C _T are the cash flows from operation at year 1, 2,…, T and r _1, r ₂,…, r _T is the discount rate at year 1, 2,…, T (Grinblatt and Titman 2002).

In the screening period before a proposed PAT strategy is accepted and an implementation is initiated, the NPV will be an estimate since the investments and gains can only be estimated. It is important that the NPV is as realistic as possible since it is used to decide whether an implementation of a PAT strategy is initiated. During a project, more details become available and the NPV will be more accurate, because at the time close to the implementation of analysers and other hardware the investments and other costs are known .

A high positive NPV indicates that the business value is high. The economic driver for implementation of a PAT strategy is there and implementation of the proposed PAT strategy may continue. The implementation of the strategy has to be accepted by the supply chain organisation, which is investing the money.

12.2.2 Organisation

Implementation of PAT obviously requires financial resources for investment in new equipment, but it also requires sufficient time resources and the people involved need the right competences. This fact is sometimes neglected, but missing focus and resources results in a high risk that an implementation will fail. Commitment from the supply chain organisation and management at the production site is crucial to a successful implementation.

The implementation of a PAT strategy should be managed by a project group, which is coordinated by a project manager. The project group should at least consist of people with competences in the production process, at-/on-/in-line analysers , laboratory analysis, multivariate data analysis , IT and automation.

Besides the competences within the project group, the following people should be available during an implementation, but they do not necessarily need to be in the project group: people with competences in contract writing and purchase to ensure a good agreement when investing, technicians from production to ensure knowledge sharing related to the manufacturing process and involvement during implementation, a maintenance department to ensure an easy handover, laboratory technicians to ensure the use of right reference methods and maintenance of the analyser and the supplier to ensure support during implementation.

Involvement of the technicians from production and laboratory in an early stage is an advantage, because after implementation, they are the end users and it will be a part of the work to establish their ownership of the new techniques and strategy.

The establishment of ownership and commitment in an organisation is important, and within a project organisation, the steering group has a central role in the work to secure commitment. The steering group consists of representatives from the supply chain organisation and/or management at the production site. The members of the steering group must have the authority to make superior decisions in relation to the project, because it is their responsibility to follow up on the progress of the project and to make superior decisions concerning the project, e.g. if changes are to be made to the purpose or description of a project. It is also the responsibility of the steering group to allocate enough and relevant resources to the project.

12.2.3 Implementation of an Analyser

A positive business value for a given project and a project organisation in place are fundamental for initiation of a project. After project initiation, the more technical part starts. The first part of a project is usually implementation of one or more analysers. The implementation of an analyser involves several steps. These steps are described in this part and the focus will be on how to ensure a successful implementation and the challenges that can appear. The steps are:

• Set up requirements for the analyser

• Do a screening of the market and assess one or two relevant techniques

• Approve the investment

• Install and develop method

• Validate the method

• Carry out maintenance

12.2.3.1 Requirements for the Analyser

The investment in a new analyser requires specific knowledge on what the purpose and economic goal are for the implementation of an analyser. Without the knowledge, it can be difficult to find the optimal analyser. Based on the knowledge, it is defined which data the analyser is going to provide to reach the goals and requirements for the analyser should be defined. The requirements can be defined based on questions like the ones listed below:

• What is the purpose of the analyser? For example, is it determination of the content of a quality attribute, like fat, protein and moisture content in cheese, or is it end-point detection, like detection of when a fermentation has ended?

• Should the measurement be at-line, on-line or in-line? What measurement interval is needed? Which type of sample needs to be measured; is it a liquid, a slurry, a powder, a solid?

• Which product characteristics need to be measured? For example, is it physical characteristics, like particle size of a powder; sensory characteristics, like the taste of a cheese; or chemical characteristics, like moisture content in cheese? What ranges or concentrations are the characteristics normally in? What accuracy and precision is needed? What are the accuracy and precision of the reference method?

• Where on the production equipment is the analyser going to be installed; e.g. is it in a pipe or a vat? What is the temperature of the product when measured? Is the product corrosive? Are there any special characteristics of the product or process, e.g. if the measurements are done in-line, what is the pressure in the pipe or vat?

• Is there a risk of segregation or a concentration gradient within the sample? This can affect the decision on if the measurements should be done in reflectance, transflectance or transmission.

• Which data from the analyser should be available for data storage, e.g. calculated concentrations and/or the spectra?

• How are the data going to be used in the production, e.g. for manual or advanced process control? Does the analyser need to be connected to the production server and IT network? If yes, how does it need to connect—through OPC, MODBUS, 4–20 mA, etc.? How should the data be available for the technicians in production, e.g. as a number or time series plot? How do the data need to be stored?

• How much maintenance can be accepted (maximum time to be used)? Should it be possible to control the instrument remotely?

The answers to all of these questions will result in a list of requirements for the analyser. The answers to the questions should focus on what is needed and NOT what would be nice. The list of requirements is then used when screening the market for potential solutions.

12.2.3.2 Screening of Market

A thorough screening of the market before the decision on which analyser to invest in is of high importance . The purpose of the screening is to collect enough information on possible techniques to be able to decide which technique best complies with the requirements. Information is found through a literature search, a search on the Internet, seminars, by talking to people from your network and by having meetings with suppliers.

Be aware that the screening should be wide in the beginning. Do not focus on only one technique or one instrument; for example, the near infrared spectroscopy (NIR) is widely used in many applications within the food industry, like measurements of moisture and protein content in bread, cheese , flour, meat, milk powder and pasta (Osborne et al. 1993) . In spite of that, NIR and other spectroscopic techniques are not always the right choice. For some applications, another and sometimes more simple technique is the optimal one, for example, measurements of salt by the use of conductivity measurements. Increased complexity for an analyser can lead to use of more resources for method development and validation. It can be a challenge to find the optimal technique, but the choice can influence whether or not the implementation of the PAT strategy will be a success. Since the analyser is to be implemented into a production environment and people without expert skills in sensor technology and multivariate data analysis are to handle the analyser, it is more important that the analyser is easy to use and robust, does not require a lot of maintenance and that data from the analyser is easy to access instead of choosing an analyser with outstanding accuracy and precision .

For applications where there is no obvious choice of analyser after a screening, it can be a challenge to decide on which technique to invest in. In such a case, it can be an idea to borrow or lease the analyser(s), which seem to be the most relevant one(s), and make screening trials in the laboratory. Trials in the laboratory can give an indication whether the analyser can comply with the requirements .

12.2.3.3 Approval of Investment and Contract Writing

The next step is to get approval of the investment in the analyser decided upon. The investment process is different for the individual companies, but usually it is the project sponsor that needs to approve the investment and the project plan. The investment that needs to be approved should at least cover the analyser, sampling equipment, data collection system, the cost of changes to the production equipment, cost for taking calibration samples and doing reference analysis and cost of training courses in calibration and maintenance of the analyser.

After approval of the investment, a contract between the production site and the supplier should be signed. A detailed contract is normally an advantage for both parties, because the responsibilities during the delivery and installation will be outlined. The contract should cover:

• A description of the aim and functionalities of the analyser, e.g. why is the analyser installed and how is it used?

• A description of the technical solution including installation instructions and timetable for the installation steps with responsibility information. Responsibility is shared between supplier and buyer depending on task.

• A description of the performance requirements, e.g. requirements for measurement accuracy and precision, handling of analyser, data validity, maximum level of unscheduled down time, safety and cleaning. It is also important to be aware if there are any limitations for the analyser.

• General issues like period of guarantee, conditions of guarantee, insurance, payment, service and milestones for the installation including factory acceptance testing.

The contract negotiation can be a challenge, but if the details are not all discussed and agreed upon between the supplier and buyer, there might be bigger challenges during the installation if something deviates from plan and responsibilities are not defined. After signing the contract, the order can be placed.

12.2.3.4 Installation and Calibration

The physical installation of the analyser, the method development and calibration is the next step when the analyser is delivered . The challenges that can appear during the physical installation depend on whether the analyser has to be used at-line or on-/in-line. An at-line installation usually only requires some space in the laboratory or operator room close to the production. On- and in-line installations require changes to be done to the production equipment (vats, pipes, etc.) and are usually more complex. One of the things to be aware of is to ensure that the installation of the analyser comply with the hygienic requirements in the production.

For both at-line and on-/in-line installations, sampling equipment for the calibration and validation samples has to be installed. Sampling needs to be performed in a way where the samples taken are representative. For calibration purposes, the samples taken need to be of the material measured by the analyser at a given time. For example, for flows in pipes and for more heterogeneous material, as seen in the food industry, sampling is a challenge. Sampling errors can be up to 1000 times larger than the analysis error (Gy 1998, 2004), so sampling is an important issue to focus on before implementation.

The physical installation of sampling equipment and the analyser may also require changes to the electrical system. The installation of the analyser and sampling equipment is based on recommendations from the equipment supplier, and it is recommended to follow these instructions closely to avoid performance issues. If any performance issues appear in spite of everything, the supplier should be able to solve the problem, since the instructions from the supplier are based on their experience and knowledge on the given application.

For some on-/in-line installations, it can be a challenge to find the optimal process interfacing (e.g. flow cell, probe head and probe angle) due to a difficult product matrix. In such cases, trials performed in collaboration with the supplier can be an advantage.

The physical installation should be completed by an approval from the supplier to ensure that the recommendations are followed.

Calibration can be initiated when the physical installation is approved, but before calibration samples are taken, calibration procedures should be in place and the responsibility for every step in the procedure should be addressed. The following should be defined and described:

• How often are calibration samples taken?

• How is a representative calibration sample taken?

• Who is responsible for taking samples?

• How is the sample handled after it is taken and until it is analysed in the laboratory? And who is responsible?

• Which reference method needs to be used and what is the exact procedure to follow?

• How is the calibration data reported?

All of the above are described in the calibration procedure and technicians in both the laboratory and production need to be trained in the procedures. If no proper calibration procedure or training is available, there is a risk that samples are taken or handled in a wrong way or that a wrong reference method is used, which can lead to a calibration with unacceptable accuracy and precision or to a long calibration period. The training of persons who have no previous knowledge within analysers, statistics or multivariate data analysis can be a challenge, but it is one of the most important tasks to succeed in. The persons who are to take the samples and do the reference analysis have to understand the importance of the procedures. During preparation of the calibration procedure, one should be aware that the more advanced a procedure is, the more errors or problems it can cause during the calibration period.

Calibration of an analyser is based on results from a reference analysis performed in the laboratory, and to achieve a good calibration it is important that the right reference method is used. The uncertainties of the reference analysis will influence the accuracy and precision of the calibration.

Calibration of an analyser requires samples, which contain varying amounts of the component of interest. Samples used for calibration should be taken in a way where they will cover the area of normal variation within the production. One can be tempted to take many samples from one production run and make a calibration based on these samples, but the calibration will not be robust enough to handle batch-to-batch variation. Calibration samples taken need to cover:

• All the products included in a given calibration

• Process variation for the component of interest

• Batch-to-batch variation, e.g. due to changes in raw materials

Variations in the production need to be included in the calibrations for the at-/on-/in-line analyser.

At the beginning of the calibration step, it can be a good idea to have a limited number of people involved in the calibration procedure. This is due to the fact that with many people involved the risk for mistakes to happen will be higher. When the calibration procedure and routines for one or two technicians in the laboratory and production are successfully worked into the daily routines, more people can be introduced and trained.

12.2.3.5 Method Validation

Validation of the installation and calibration is the final step before the implementation is ended and the analyser is included into the maintenance procedures at the production site. Validation of the installation covers evaluation of whether, based on the experiences from the calibration period, there is anything in the practical handling of the analyser that is not appropriate. It also covers an evaluation of whether the technical requirements are fulfilled.

Validation of a calibration covers the evaluation of whether the required accuracy and precision is reached. This validation is done based on results for validation samples, which are sampled and analysed using the exact same procedures as the calibration samples.

12.2.3.6 Maintenance

Maintenance of the implemented analyser is included into the normal maintenance procedure and is done by the personnel at the production site. The maintenance is divided into two areas, hardware maintenance and calibration maintenance. Hardware maintenance is done to ensure that the consumable parts, e.g. the light source, are changed in time to ensure that the analyser keeps working. Many analysers have the possibility to set up automatic tests for performance checking of the instruments, and if these tests fail, alarms are activated.

Calibration maintenance is done to monitor and ensure the accuracy and precision of the method. Calibration maintenance includes making control charts for the reference analysis in the laboratory to ensure that the reference analysis is in control and can be used. It also includes making control charts for samples taken on a regular basis to monitor the prediction error (calculated as the difference between the predicted value from the analyser and the reference value from the laboratory). An example of a control chart is shown in Fig. 12.2.

Fig. 12.2

Control chart for monitoring the difference between predicted and reference value

If relevant, the calibrations are recalculated or updated. Recalibration can be necessary due to drift in the analyser signal or changes in the composition of the samples.

Maintenance is important for the continued success of the implementation of a PAT strategy. For the maintenance to be successful, it is important to have focus on resources and competences. The managers at the production site have the responsibility to solve this challenge but it might be an idea to appoint a person, who is responsible for the maintenance and for giving the managers warning if challenges on competences and resources appear .

12.2.4 Control Strategy

Process optimization and process control based on at-/on-/in-line analysers can be more or less automatic. Three different strategies can be used.

• Manual control: The results from the analyser are logged into the data collection system in the production in real time and shown on the production technicians’ screen in the control room. The production technicians will take action if necessary, based on these results and their own experience.

• Advisory system: The results from the analyser are logged into the data collection system in the production in real-time together with all the other process variables. Based on the real-time data and a process model calculated from historical data, the advisory system will calculate the optimal settings for the process set point and advise the production technician on which changes to make if necessary.

• Automatic control system: The results from the analyser are logged into the data collection system in the production in real time together with all other process variables. Based on the real-time data and a process model calculated from historical data, the automatic control system will calculate the optimal settings for the process set point, and if necessary, the set points will be changed automatically.

The control strategy to choose will be dependent on the IT structure available, the results of the business value calculation , the size of the production at the production site and on the available resources and competences. The control strategy should be selected based on requirements and not because it is the most simple or advanced strategy.

A manual, advisory or automatic control system, fully or partly based on multivariate data analysis , depends on the relevant process data and data from at-/on-/in-line analysers being logged into the system in real time. Without real-time data collection and data handling systems, a control system is impossible to build into production.

Worldwide, there are many different suppliers of data collection and data handling systems on the market, and which system is the most optimal depends on which other control systems there are on a given production site. Compatibility between the different IT systems on a production site is very important.

12.2.4.1 Data Collection and Data Handling System

Data collection and data handling is critical for whether an advisory or automatic control system will work or not. For the control strategies to succeed, all the relevant process data and data from at-/on-/in-line analysers need to be logged. Process knowledge is important when deciding which data could be relevant and not. All process variables that can influence the final product should be logged. It is better to log all the process data than to log too few data, because the nonsignificant process variables can easily be excluded from the process model during the modelling stage.

Whether the advisory or automatic control system is built for a single process unit or for a total production process, the steps needed to define relevant data are the same. A list of the process variables that are already logged should be made and the different steps in the production process should be analysed, based on this list, to decide if more process variables need to be logged. Afterwards, a plan for how to implement the data collection should be made. In the same way as for a new analyser, a list of requirements should be made for the data collection and data handling system. The other steps like screening of the market , approval of investment and contract writing, installation and validation are just as important for a data collection and handling system as for a new analyser.

Since the process data and data from the at-/on-/in-line analysers are crucial for an automatic control or advisory system, the data always need to be valid and available. In some data collection and data handling systems, it is possible to register if an analyser fails and no data are logged into the system. In such a case, some systems can also give an alarm to raise awareness of missing data, by either sending an email or an SMS. An alarm makes it possible to solve the problem as soon as possible.

Data collection, data handling and IT communication between the different IT systems in a production is crucial to a running PAT system, and focus on IT and IT risk management is important to ensure a well-functioning system.

A benefit from implementation of an extensive data collection and data handling system is that, from the data, there is complete traceability from raw material to final product. All records for a given product will be stored within the data collection and data handling system.

12.2.4.2 Process Modelling

The basis of an advisory or automatic control system is the process model, which can be based on multivariate data analysis , first principle models or a combination. The process model relates measurements of raw material and process variables to the outcome of a process or process step and extracts information on how all the variables are behaving relative to one another. This will contribute to more process knowledge.

Before process modelling is initiated, it should be ensured that the logged data for the process variables and from the analysers are valid and of high quality. Without high-quality data, there is a risk that the process modelling will fail. The task requires resources to validate the process data and it can be a challenge.

In the same way as for the calibration of an analyser, the process data included in the modelling need to cover all the products included in a given process model, batch-to-batch variation and sufficient process variation. A data set with sufficient variation can take months to collect, particularly if variation during normal operating conditions is not big enough to reveal obvious cause–effect relations. In such cases, step tests with planned and supervised changes to the process settings can usefully be made. The challenge is to define how big changes to the process settings can be allowed without the risk of producing product outside specifications.

12.3 Case Studies: Process Optimization Based on Quality Attribute Measurements

Quality attributes for dairy products can be both the chemical composition of a given product like protein, moisture and fat content, and the sensory quality attributes like taste, smell and consistency. The quality attributes of products or semimanufactured products can be measured using PAT tools. Table 12.1 shows some examples of PAT tools employed for various operations within the dairy processing industry. The asterisks in the table indicate how widely the techniques are implemented.

The case studies in this chapter will focus on cheese production and describe the goals, advantages and challenges of implementing a measurement system for moisture in semihard cheese and a sensory quality assurance system for cream cheese.

12.3.1 In-line Moisture Measurement for Semihard Cheese

Measurements of the quality attributes of products are important for process control , and for semihard cheese, one of the quality attributes is moisture content. The variation in and level of moisture in cheeses influences the yield . Control of the moisture content in cheese is therefore also a part of optimization of the yield in cheese production. In-line measurements of the moisture content make it possible to control the cheese vats and pressing system to reach a predefined moisture content. Control based on the measurements results in a reduced batch-to-batch and in-batch variation and gives good and stable quality and a higher yield.

The requirements that were defined for an analyser to measure moisture content in semihard cheese resulted in the selection of a technique based on microwave technology. The advantage of this technique is that some available microwave sensors can measure in transmission through a cheese block, which is important because there is typically a moisture gradient within the block. A cheese block can be up to 140 mm high. The other requirements were, as described in Sect. 12.2.3, related to accuracy and precision of the measurements, data storage, data communication and the process conditions at the measuring point. In the given case, a specific requirement related to cleaning was defined, because the analyser was installed in a place where CIP (clean in place) was performed.

The instrument was delivered, installed into the production and approved by the supplier.

12.3.1.1 Changes of Routines in Production and Laboratory

The decision to implement the in-line moisture analyser at a production site resulted in the need for new routines in production and in the laboratory. A routine for how calibration samples are taken during production needed to be introduced and in the laboratory there was a need for new routines concerning reference analysis. The change in routines can be a challenge and can also require new sorts of competences.

Changes in routines can also be caused by the product being measured and sampled. In the case of semihard cheese, one of the challenges in the calibration work was the sampling. The sampling needs to be performed in such a way that the samples taken from a block of cheese of minimum 20 × 20 × 9 cm is representative of the given cheese block and taken without changing the composition of the sample. There is a risk of changing the composition of a sample of fresh semihard cheese because when the sample is taken the cheese is cut, and when a fresh semihard cheese is cut, whey starts to drain from the cheese. The whey drainage results in a decrease in moisture content in the sample taken to the laboratory for reference analysis. To ensure that the calibration for the analyser is as accurate as possible, specific routines were implemented for how to take the calibration samples and how to handle the sample prior to reference analysis. The calibration samples were taken at the production line, and as soon as the sample was cut from the cheese, it was placed into two bags and transported to the laboratory right away. In the laboratory, the sample and the small amount of whey that had drained from the cheese into the bag was mixed prior to the reference analysis to ensure that the measured moisture content was not affected by the whey drainage.

It is not enough just to implement a new routine. Another important part is to have the production and laboratory technicians to understand why it is necessary that they take the sample in the way described and why they must use a given reference method. For production personnel who are not familiar with on-/in-line analysers and do not understand why analysers need to be calibrated, it can be hard to understand why they need to follow routines that sometimes are time consuming and difficult to carry out. Project managers for an on-/in-line implementation need to be aware of this challenge and find a way to handle it. In cases where the new routines are more time consuming during the calibration period, it is also important to get management support to have the necessary resources. If the necessary resources are not made available, there is a risk that the calibration period will be prolonged or that the calibration obtained will not be satisfactory and cannot be used for control purposes.

The calibration work should not be initiated until the routines are in place.

12.3.1.2 Calibration of the In-line Moisture Analyser

Implementation of the in-line moisture analyser required variation in the moisture content of the cheese. As described in Sect. 12.2.3, the calibration samples need to cover all the products included in a given calibration, process variation for the component of interest and batch-to-batch variation .

In a dairy, different sorts of cheeses are produced. The cheeses differ in height, size, moisture, fat and protein content. Based on these variables, the cheeses can be grouped according to similarities. Some types of cheese are sufficiently alike so that they can be included in the same calibration to keep the number of calibrations to a minimum. Less number of calibrations results in less calibration maintenance work.

Calibration of the in-line moisture analyser (using microwave transmission) for each group of cheeses requires that samples are taken for every sort of cheese included in the group and that the samples taken vary in height and moisture content. This seems like a fairly easy task, but within a production environment challenges appear. The planning of when samples are taken can be difficult.

In a cheese plant, the production is planned from week to week and the production is highly flexible. The flexibility and the fact that different sorts of cheese are produced results in a production plan that varies from week to week, and during a day different sorts of cheese are produced. It is rarely the case that one sort of cheese is always produced on a specific day of the week. The flexibility in the production plans is a challenge when the people at the dairy have to schedule the resources for the calibration work.

The production at a dairy typically runs for 24 h a day, and sometimes the product of interest is produced during the night when no laboratory technicians are on site. In other cases, the product of interest is planned for production during the day but a technical problem then arises during production and the production is delayed. Such cases can cause calibration work to be disrupted from time to time .

12.3.1.3 Benefits of Implementation of an In-line Moisture Analyser

Even though implementation of an in-/on-line analyser takes time and requires the right competences and resources, there are important benefits.

The measurements from the in-line moisture analyser resulted in new insight into the cheese-making process. The cheese-making process consists of multiple process steps. Cheese milk is added into a cheese vat and then culture and rennet is added to ensure acidification and coagulation. After the coagulation of the cheese milk, the coagulum is cut and stirred. Whey is removed from the cheese vat before the cheese curd is pumped into a pressing system, where it is pressed and formed into cheese blocks. The cheese blocks are then brined and stored.

The in-line moisture analyser is installed between the pressing and brining step as a measurement of a critical product quality attribute for the fresh produced semihard cheese. The measurements of the moisture in the fresh cheeses revealed that the moisture content in most cases decreased within a batch; an example is shown in Fig. 12.3.

Fig. 12.3

In-line measurements of moisture in cheeses for five different batches

The first cheeses within a batch have higher moisture content than the last cheeses within a batch. The measurements confirmed the intuitive knowledge about the process. It had never been possible to document this in detail before because this kind of analysis would have required a lot of manual sampling and reference testing in the laboratory, which would have required many resources.

The benefit of the in-line moisture analysis is that the variation within the production can be documented, and based on the measurements it is possible to make changes to the relevant process variables to reduce the systematic decrease in moisture content within a batch of cheese.

An example of in-line moisture measurements for two batches where changes were made to the relevant process variables and three batches without changes is shown in Fig. 12.4.

Fig. 12.4

In-line measurements of moisture in cheeses including batches where changes were made to the process variables to reduce the decrease in moisture content. Light blue and purple lines represent batches with changes. Blue, green and red lines represent batches without changes

The in-line measurements for the batches where changes were made to some relevant process variables show that the systematic decrease in moisture content has been reduced compared to the batches without changes. These results show one of the potentials for process optimization based on the in-line moisture measurements .

The process optimization and process control based on the in-line moisture measurements can be done in three ways, manual control based on real-time feedback to the production technicians or control based on advanced process models for an advisory system or automatic control system. The implemented control strategy will vary from dairy to dairy dependent on the result of the business value calculation , the size of the production at the dairy and on the available resources and competences at the given dairy. The in-line measurements can be used for optimization of the internal batch variation by subsidiary optimization for the most relevant process step, but in the case of the advanced process control the measurements can also be used to build a process model, which is used to predict the optimal settings for the overall process set point to ensure a specified moisture content in the cheese produced.

12.3.2 Sensory Quality Assurance System for Cream Cheese

The key quality attributes for cream cheese are sensory attributes related to the consistency, taste, smell and appearance of the cheese. The sensory attributes need to be measured to be able to optimize the production according to these attributes. Sensory attributes can be difficult to measure with at-/on- and in-line analysers, but even though it is a challenge, there is a potential in optimization of the sensory quality of food products . The economic gains will be a reduction in amount of product outside specifications and thereby in the amount of rejected product.

Instead of using at-/on- or in-line analysers , a sensory quality assurance system was developed with the focus that the assessments could be used in an advanced control system. In the same way as for an instrumental analyser, as described in Sect. 12.2.3, requirements were defined for the sensory quality assurance system both within the performance area and for data collection . The assessments needed to give interval data and to be on a linear scale instead of nominal data and category scale. The assessments also had to focus on possible deviations from product specification for optimum product instead of assessing total characteristics like smell, taste, consistency and appearance. Some possible deviations could be that the cream cheese is too soft or that there are small grains of fat. An assessment of all possible deviations from specification for a product makes it possible to correlate the results to how the production was run, in order to gain process insight, to find the critical control points and to make it possible to control the production to produce a product with a predefined quality and reduce the amount of rejected product.

12.3.2.1 Principle of the Sensory Quality Assurance System

In the sensory quality assurance system, the assessment is divided into two steps. The first step for the assessors after they have tasted the product is to decide whether the product is approved, approved with a remark or rejected, based on the product specifications. “Approved with a remark” means that the product is within product specifications but with small deviations from optimum product. The conclusion from the first step of the sensory assessment is whether the product can be sold.

The second step in the assessment relates to process optimization. Optimization of the process to gain an improved sensory quality is focused on the ability to control the production to reduce the amount of product categorized as “approved with remark” or rejected. To reduce this amount, it is necessary to assess if a given product deviates from specifications for the optimum product, and if it does, then to assess by how much it deviates. Results from the second step in the assessment can then be used to investigate what, in the production process, caused the deviation. For every product type, a list of possible deviations from optimum product is defined—see example in Fig. 12.5.

Fig. 12.5

The sensory quality assurance system

The degree of deviation is assessed on a line scale, where it is important that the line scale is unstructured. An unstructured line scale has no numbers on the scale, because such numbers can be seen as anchor points on the scale and the assessments will then usually be placed at these points. The only anchor points on the scale are None and Most, for which the assessors have definitions. Assessments on a line scale also give normally distributed data, which is preferred. Normally distributed data would not be achieved if the assessors were just to place a cross on predefined points. Without normally distributed data, it is not possible to make the relevant data analysis, like, for example, ANOVA.

12.3.2.2 Training of the Assessors and the Challenges

Implementation of a sensory quality assurance system includes training of assessors. The assessors need to go through training to be able to manage the assessment system, just like an analyser needs to be calibrated. Training of the assessors is as important as calibration of at-/on-/in-line analysers and it takes time and effort. The assessors need to get the right competences.

During training, the assessors need to understand the definition of maximum degree of deviations for a given product and have to assess reference samples taken from the production. The reference samples need to have various degrees of deviation. For every reference sample, the assessors need to agree on which area the sample is placed on the line scale, which is done during a discussion between the assessors based on the product specification. The training can be challenging because every assessor needs to agree on the decision. It is necessary to be aware that in a sensory quality assurance system the analyser is human and not a mechanical instrument, which can be reset or recalibrated. In some cases, this requires that work needs to be done to change the mindset of an assessor, who may previously have worked with other assessment systems.

The purpose of the training is to ensure that the assessors can repeat their test results on the same sample over time, that they can separate products with different degrees of deviation from optimum product and that they can rank the products based on degree of deviation.

When this is documented, the system can be taken into use at the production site. To ensure and to document that the assessors can repeat themselves over time, a standard sample or a sample from production, which is used repeatedly, is assessed when products from normal production are being assessed. To document that there is agreement between the assessors, it is important to have replicates, so more assessors need to assess the same products from normal production. This can result in changes in routines in the daily work, because the assessors and their managers need to plan when the assessments are performed.

As for at-/on-/in-line analysers, calibration , validation and maintenance of the sensory quality assurance system and the assessors are of high importance. For a sensory quality assurance system, it requires that training sessions be carried out with regular intervals. For a successful implementation and continued use of a sensory quality assurance system, resources are crucial, but as for calibration and maintenance of an analyser, it can often be a challenge to get enough resources because the assessors also have other tasks to solve during a working day. The managers at the production site have the responsibility to solve this challenge.

12.3.2.3 Benefits of Implementation of a Sensory Quality Assurance System

Implementation of a sensory quality assurance system requires time, training of the assessors and the right competences and resources, but there are very important benefits to gain from the work.

The assessments in the sensory quality assurance system are performed on a computer, and the data are saved into the data collection and data handling system together with relevant process data. The assessment results from the sensory quality assurance system can be analysed by statistics and multivariate data analysis , which makes it possible to correlate the results to how the production was run. This will generate more process knowledge. The assessments could be included into a process model and into a control system to make it possible to control the production to ensure production of a product with a predefined quality and reduce the amount of rejected product.

Besides the economic gains, the results from the assessments can be used as part of a quality management system at a production site. As stated in the previous section about training of the assessors, it should be possible to document that the assessors can repeat themselves over time and to document that there is agreement between the assessors. Since the results from the assessments are automatically saved into a data collection and data handling system, the data will always be available and documentation on the performance of the assessors can be extracted. This documentation can then be used when an audit is performed at the production site.

12.4 Summary

The intent of this chapter was to present some food industry perspectives on the advantages and challenges when implementing a PAT strategy .

The driver behind implementation of a PAT strategy must be an economic gain obtained by improved yield , increased product quality, reduced amount of product outside specifications or reduced product costs. Besides the economic gains, the implementation of PAT results in other benefits like more process knowledge and understanding. As part of the implementation of a PAT strategy, a data collection and data handling system is implemented which can give the advantage of complete traceability in a production site.

Management commitment and support is crucial to ensure a successful implementation of PAT, because the necessary focus, resources and competences have to be available throughout a project and after implementation. It is important to be aware that the investment made during an implementation of a PAT strategy is not only in physical instrumentation (e.g. in an analyser) but also in resources and competences for calibration, validation and maintenance of the analyser and maintenance of a data collection and data handling system. The steps and challenges to overcome when implementing an analyser were described. The list of steps is not necessarily complete, but hopefully it will be an inspiration and basis for further discussion at sites where a PAT strategy is going to be implemented.