Knowledge Provenance Management System for a Dropwise Additive

The Knowledge Provenance Management System, KProMS, can capture the complete provenance of the data, information, and knowledge of a structured ...
0 downloads 0 Views 8MB Size
Article pubs.acs.org/IECR

Knowledge Provenance Management System for a Dropwise Additive Manufacturing System for Pharmaceutical Products Elçin Içten,* Girish Joglekar, Chelsey Wallace, Kristen Loehr,† Jennifer Sacksteder,† Arun Giridhar, Zoltan K. Nagy, and Gintaras V. Reklaitis School of Chemical Engineering, Purdue University, West Lafayette, Indiana 47907-2100, United States S Supporting Information *

ABSTRACT: The Knowledge Provenance Management System, KProMS, can capture the complete provenance of the data, information, and knowledge of a structured activity by modeling the details of the associated data generation steps of that activity as workflows. Its unique workflow representation captures relationships between the processing steps, material and information flows, and data input and output. In this paper, we demonstrate the use of KProMS to manage and analyze the experimental data of an innovative system for manufacturing drug products using dropwise additive manufacturing. Dropwise additive manufacturing of pharmaceutical products (DAMPP) uses drop on demand printing technology for depositing various drug formulations onto edible substrates. DAMPP requires and generates a range of data types, including camera and IR images, spectra, and numerical parameter values, both of real-time and off-line natures, and thus provide a rich illustration of KProMS capabilities to serve as knowledge management framework.

1. INTRODUCTION Dropwise additive manufacturing of pharmaceutical products (DAMPP) is a novel technology for small scale, distributed manufacturing of individualized solid oral dosage forms. The DAMPP process utilizes the drop on demand (DoD) printing technology for predictable and highly controllable deposition of active pharmaceutical ingredients (API) onto an edible substrate. The feasibility of DAMPP has been successfully demonstrated for melt-based1 and solvent-based2 formulations. These feasibility studies included a variety of solvents and polymers as carriers, active pharmaceutical ingredients (API), and substrates such as polymeric films and placebo tablets. If desired, the drops can also be deposited into capsules as final dosage form. This manufacture method has tremendous potential in personalized medicine because, through a combination of drop size and number of drops, the dosage can be precisely and reliably controlled to match the prescribed amount for an individual. In addition, the “batch size” can be as small as the number of tablets or drops required for one patient and the transitions from one dose to another made with minimal or no waste in very short time. If the DAMPP facility were used to manufacture drug products at a commercial scale, then the US Food and Drug Administration (FDA) would require detailed information about the manufacturing process to support a filing for FDA approval as well as for reporting on quality compliance during the actual manufacturing of specific products. For example, if one set of drops deposited onto a placebo tablet would constitute one dose, then data would be required to prove that the dosage © XXXX American Chemical Society

produced indeed meets the required amount of active substance. In addition, data for other critical quality attributes, such as dissolution rates, would also be required. If the facility were used by a compounding pharmacy or clinic for production of a dosage for immediate use by patients, then too a record of the production to the dosage would be needed in order to document that the dosage produced met specifications. In either case, a system is required to record and make accessible the data provenance of each and every dosage produced. To address this need, the DAMPP system has the instrumentation to collect a range of process variables for process monitoring and control. Additionally, it is integrated with the Knowledge Provenance Management System, KProMS,3 for the purpose of managing all information generated during a run.4 For a typical run, KProMS is used to set up each production run and store the data generated during that run, thereby providing a single point of information access. In this paper, we demonstrate the use of KProMS in managing all aspects of running a DAMPP system. First, a detailed description of the main aspects of operating the DAMPP installation is given, followed by the description of the steps in creating an integrated application with KProMS. Finally, the use of the integrated system for typical data retrieval, analysis, and predictive functions is described. Received: March 16, 2016 Revised: July 16, 2016 Accepted: July 27, 2016

A

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX

Article

Industrial & Engineering Chemistry Research

2. DROPWISE ADDITIVE MANUFACTURING PROCESS DESCRIPTION AND OPERATION The prototype DAMPP system consists of a material reservoir, precision positive displacement pump, xy-staging, a hot-air-based heating system, online imaging and sensing, and temperature, pump, and stage controllers. The material reservoir, pump, nozzle, camera, substrate, and staging are labeled in Figure 1 with numbers 1−6, respectively. Using this system, different

with the material properties, i.e., viscosity, surface tension, and density, are the main factors that affect reproducible and precise drop formation, and thus drop size and dosage amount. The details of drop formation with the DoD technology are discussed further in section 5. An image of each drop is taken after it is ejected from the nozzle, while in flight, using the online imaging system. The size of each drop is monitored using real-time image analysis. Ability to adjust and verify the drop size and thus the dosage amount in each dose enables reproducible production of individualized dosing. Drop positioning is accomplished by positioning the printer nozzle above the desired location on the substrate before drop ejection. The xy-staging and synchronization logic assure precise drop positioning on the substrate while printing and enable layering of multiple drops of the same or different drugs. The temperature of the deposited drops is controlled indirectly by controlling the temperature of the substrate using a Peltier device with a PID controller, which is placed on the xy-stage. This allows control of nucleation and crystallization phenomena. Since the crystallization temperature profile has a strong effect on product solid state characteristics, influencing the morphology and the dissolution properties and hence the bioavailability of the drug, precise control of the drop solidification process occurring on the substrate is crucial. As a drug product manufacturing system, the DAMPP process must be operated so as to ensure and document that critical product quality specifications are met. Thus, it requires precise monitoring and control of the manufacturing operations as well as the capture and organization of all of the information associated with these operations. The use of DAMPP for a new product formulation generally requires execution of two experimental phases: determining the range of conditions under which good quality drops are formed consistently for that formulation, followed by operation of the manufacturing system under the most desirable operating conditions in that range. In the case study reported here, the use of KProMS to support both phases is addressed. During DAMPP operation, the input settings associated with all process units are specified through the Labview operator interface. The input file contains all the input parameter settings, which are required to run the automation program. As the first step, a selected drug formulation is filled into the materials reservoir. The second step consists of the IVEK pump,5 whose operating parameters, pump rate (RPM), pump displacement, and volume strokes, must be specified. Using the precision IVEK pump, the material is ejected through the nozzle and deposited on the substrate. The motion sensors detect the drop, and the camera automatically captures an image of the falling drop. Using real-time image analysis, the volume is calculated for each drop, and the corresponding dosage amount is calculated from the known formulation density and drug concentration.6 The synchronization logic enables the deposition of the drops on the substrate with predefined x−y coordinates, which allows creating a grid-like deposition pattern as well as multilayered drug formulations. After deposition, the substrate temperature controller is used to control the cooling temperature profile of the drug deposits.7 An infrared (IR) camera is used to monitor the spatial temperature distribution within the deposit.8 Both the solidification temperature profile and the formulation properties can affect the bioavailability of the drug products. Therefore, postprocessing steps could include different quality control tests such as crystallinity measurement or dissolution testing. At the end of

Figure 1. Dropwise additive manufacturing system (1, reservoir; 2, IVEK pump; 3, nozzle; 4, camera; 5, substrate, tablet holder; 6, xy-stage).

drug formulations can be processed where the API is dissolved in a solvent- or melt-based formulation. Depending on the desired final solid state form of the API, different carriers, such as polymers, can be used in the formulation. The key characteristics affecting the solid state form of the drug products are the nucleation and crystallization growth rates of the API during solvent evaporation or melt cooling. Polymers are added to the printing material not only to help control drug solid state properties, but also to establish formulation composition and material rheological properties. The material rheological properties of the formulation depend on the surface tension and viscosity values of different polymers used. Thus, the material source, properties, and composition are important parameters affecting the product quality. In order to process melt-based formulations reproducibly, temperature control is established on the whole process including reservoir, pump, tubing, and nozzle. Controlling reservoir temperature tightly allows dissolving the drug substance in the molten polymer below the potential degradation temperature of the drug. Controlling temperature on the processing line allows constant material rheological properties of the printed formulation to be maintained. The fluid formulation is pumped through the precision positive displacement pump with a custom controller, which allows for the adjustment of the drop size by changing the pump parameters. After passing through the pump, the drops are ejected through a nozzle, which can be of various internal diameter sizes as required by the specific formulation. Printing parameters, i.e., pump parameters and nozzle diameter, along B

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX

Article

Industrial & Engineering Chemistry Research

Figure 2. DAMPP system workflow.

flow; and information input, output, and flow. The workflow, named TB3Basic, for modeling the DAMPP system is shown in Figure 2. Each green rectangle represents a task associated with a component of the DAMPP system with the white rectangles representing subtasks carried out as part of that task. Material transfers between subtasks are shown with solid lines and information transfers as dotted lines. Yellow pentagon indicates an input material and the yellow triangle an output product. Tan rectangles denote data nodes. (Further details of specific workflow symbols are provided in Figure A.1 in Appendix.) 3.2. Management of DAMPP System. The first step in managing the information related to the DAMPP system is to create the workflow for the associated processing step as shown in Figure 2. Each production run of the system constitutes one instance of the workflow. A production run is associated with the set of drops on substrate strips/tablets produced under a given set of fixed operating conditions. The values of operating parameters are entered through conventional data entry forms defined for each task, subtask, material, and data node. By way of example, the data entry form for the pump settings are shown in Figure 3. (Additional data entry forms are shown in

a run, the LabVIEW program generates an output file containing all the process monitoring and control parameters for each drop deposited. This includes x−y coordinates, image file name for each drop, drop volume, drop center, and actual temperature values. Also, the drop images, IR images, and quality control results are saved for each run.

3. KNOWLEDGE PROVENANCE MANAGEMENT SYSTEM: KPROMS Information technologies are transforming how process knowledge is generated, managed, and analyzed. In the last few decades, the development of knowledge-based systems has received increasing attention in the chemical engineering domain.9 Recipe management systems are now in use to track data from batch chemical processes and to generate/regenerate the batch recipe for each batch.10,11 Knowledge-based systems have been developed for the representation of mathematical models for pharmaceutical product development12,13 and for other engineering purposes.14,15 Some degree of knowledge capture is provided by electronic lab notebooks and laboratory information management systems (LIMS). Laboratory e-notebooks, such as ELN,16 primarily offer capabilities for data capture in spreadsheet format, importing data from instruments and providing links to useful resources such as searchable reaction databases. LIMS, such as BIOVIA LIMS,17 have been successful in supporting discovery research for standardized processes such as high throughput screening and do offer data archiving and search capabilities. However, current products provide limited functionality in terms of capturing metadata, complex workflows, experimental designs, and the layers of activities and information, which fully document data provenance. The workflow-based Knowledge Provenance Management System, KProMS, captures the complete provenance of knowledge by modeling the details of the associated knowledge generation steps as a set of hierarchical workflows. Its unique workflow representation captures relationships between the steps as a network of nodes connected by material and/or information flows. Each data input or output step is also included in the workflow as a node connected by an information flow to the step processing or generating the data. The general framework can be used for experimental, scientific, and business workflows and manufacturing recipes.3 KProMS has been implemented to function as a HUBzero18 component and thus enjoys all of the web-based features of that environment. 3.1. Using KProMS. The use of the KProMS system broadly consists of two steps: the creation of a workflow(s) to represent the knowledge generation steps and the use of the data retained with the workflow(s) to create knowledge. A graphical Workflow Builder is used for creating a workflow using a set of simple workflow building blocks: workflow; task; subtask; material source/sink, intermediate material input, output, and

Figure 3. Data entry form for TB3Basic workflow.

Appendix Figure A.2). A total of 29 task and subtask parameters define the operating conditions for a DAMPP run, including the set points for pump rpm, number of volume strokes, and four temperature settings (reservoir, pump, material line, and substrate). The set points are used by the LabVIEW system, which controls the operation of the system. The steps in executing a run are shown in Figure 4. The PC that runs the system via LabVIEW also runs the KProMS GUI on the FireFox web browser. After setting up an instance using the KProMS GUI, an intermediate text file consisting of the parameters defined in the XParForTr data node shown in Figure 2 is downloaded on the PC in a specific folder. (An example intermediate text file is shown in Figure A.3 of the Appendix.) The virtual instrument (VI) created for the system is started, and the intermediate file is loaded into the VI. C

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX

Article

Industrial & Engineering Chemistry Research The operator confirms that the loaded parameter values are correct by observing the user interface of the LabVIEW automation program shown in Figure 5 and starts the run. The run is executed by LabVIEW via communication with the process

equipment through the serial and USB ports on the PC according to the design of the VI. The details of the LabVIEW based automation program developed for the DAMPP process have been previously reported.6 During the run, LabVIEW creates a tabular text file consisting of a fixed set of columns. Each row in the file has the following key pieces of information about each drop deposited on the substrate: drop volume computed from the processing the drop image captured by the camera, the name of the .png file which has the drop image, and the x and y coordinates of the drop on the substrate. (An example output file is shown in Figure A.4 of the Appendix). At the end of the run, the LabVIEW output file and a compressed file consisting of all drop image files are uploaded into the LabViewData data node of the workflow using the KProMS GUI. Similarly, crystallinity data and the IR data about a preselected set of drops, if applicable, are loaded in the CryData and IRData data nodes of the workflow. Thus, a completed run

Figure 4. Integration of KProMS and the DAMPP test bed.

Figure 5. User interface of the LabVIEW-based automation program. D

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX

Article

Industrial & Engineering Chemistry Research

object, in this case, since WfObjlvl1 is a data node. Thus, the parameters of that data node constitute the choice list for WfObjlvl2. Once the data node parameter is selected, the columns in the associated metadata constitute the choice list for parameter field as shown in Figure 7. As shown in Figure 7, the Drop Number column is selected to completely define the tuple identifying the x variable, and Drop volume is selected to completely define the tuple identifying the y variable of the graph. The GraphData program of KProMS draws a plot, an example of which is shown in Figure 8a, where the drop volume versus the drop number is plotted for six different DAMPP production runs 255−260. This plot corresponds to the data selection process shown in Figure 7. Runs 255−257 were performed with a pump rate of 400, pump displacement of 2.5, and volume strokes of 1. These conditions yield an average drop volume of 14.24 μL. For the runs 258−260, the displacement is 1.5, which decreases the average drop size to 9.88 μL. The KProMS system has a library of programs written in PHP to perform specific computational tasks, such as full factorial DOE, linear multiple regression, and basic statistics for columns of data such as mean, average, standard deviation, and relative standard deviation. For the DAMPP run instances, the average drop volumes, standard deviation, and relative standard deviation within each instance and between different instances are calculated. The reproducibility of drop sizes is crucial for this process, since it directly correlates with the variability in the dosage amount. For instances 255−260, the results of the statistical calculations are shown in Figure 8b,c. While this statistical analysis example is directed at postprocessing of a series of runs, statistical testing to ensure that each dosage instance meets quality limits can be implemented in a similar fashion. It is straightforward to define a quality control workflow which uses the established acceptance criteria to test the quality attributes of each dosage as it is completed so as to make a real-time determination of acceptance. This can include a virtual sensor which can relate online measured properties to quality tests such as the standard dissolution test. One such predictive model is reported in Icten et al.7

that is stored in the repository has the complete provenance of the associated run, including all process parameters and outputs along with the relationships defined by the graphical network of the workflow.

4. DATA EXTRACTION AND ANALYSIS In order to generate useful knowledge from the execution of the process workflow, data recorded through KProMS can be extracted and analyzed. A data node in a workflow can have any number of parameters, each of which has two key attributes: the name of the file in which the associated data is stored and the metadata for the data. The metadata can be predefined or derived from the file structure. In addition, each piece of information stored in the repository using KProMS is accessible with a unique tuple. For example, a task parameter is identified by the triplet (workflow name, task name, parameter keyword), a subtask parameter identified by the 4-tuple (workflow name, task name, subtask name, parameter keyword), a column in a data file by the 4-tuple (Workflow name, data node name, parameter keyword, column name), and so on. The knowledge of metadata and a unique identity for each piece of information facilitate extraction of data using context sensitive menus. A data node of type Extract Data can be incorporated into a workflow to represent data extraction step, which could be part of a scientific workflow for analyzing data. An example of a general purpose workflow to draw x−y plots is shown in Figure 6.

Figure 6. Scientific workflow for drawing plots.

The Select Data node allows specification of the sets of x−y data to extract, for single or multiple workflow instances, which in this case are DAMPP runs. In the following example, the data selection process for plotting volume of each drop from 6 different runs is described. The form for selecting data is shown in Figure 7. Typically, in a data extraction node, the data sets that are selected have matching provenance. Therefore, first, the parent template for the associated runs is selected, TB3Basic:Experimental in this case. Then, instances of that template are selected from the table created of children of the selected template. In this case 6 different process runs, 255−260, are selected. Next, the x and y axis data to be graphed are selected. The variables are selected using context sensitive menus to identify the unique tuple associated with each. The first variable in the list identifies the x axis variable and the second variable the y axis data. In this example, both drop number and volume are columns in the data file associated with the LabViewData data node of the workflow. These columns are defined in the metadata of the data node. The first three items defining the tuples associated with the x and y variables are the same, namely, Workflow template (field Workflow), Data Node (field WfObjlvl1), and Data Node Parameter (field WfObjlvl2). The choice list for WfObjlvl1 is created using the workflow template. It consists of all children of the workflow, namely, task node names, data names, raw material names, and sink names. The choice list for WfObjlvl2 is created using the WfObjlvl1

5. KNOWLEDGE EXTRACTION: OPERATING REGIME Drop quality depends on the drop dynamics which is affected by fluid properties and operating conditions. For stable drop formation of a selected formulation, the operating regimes should be characterized and the best set of operating conditions determined. This can be done empirically for a given formulation by executing a set of designed experiments. However, the results of such experiments with a variety of formulations can be accumulated over time and aggregated via dimensionless numbers to develop a more general operating region description, which can then be used for predicting likely conditions to be used for a new formulation. The KProMS system facilitates aggregating experimental data to create these alternative levels of operational knowledge. In the DAMPP system, the pharmaceutical formulation contained in the reservoir is pumped through a drop on demand (DoD) positive displacement pump. DoD technology allows generation of individual drops by means of a pressure pulse upstream of the printing nozzle.19,20 Until the pressure pulse reaches a threshold value, the liquid will be held in place by the surface tension at the nozzle. Once the threshold value is exceeded, a drop is ejected.20 With adjustment to the pressure pulse used to form the drops, the drop size can be controlled E

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX

Article

Industrial & Engineering Chemistry Research

Figure 7. Data selection through SelectData node.

Figure 8. (a) Drop number vs drop volume for instances 255−260. Basic statistics results for instances (b) 255−257 and (c) 258−260.

within a defined range.20 Different DoD printing methods, including thermal, piezoelectric, or positive displacement pump, can used to generate the pressure pulse required for drop formation. In the DAMPP, the precision pump is connected to a custom controller that allows for variations in printing parameters such as the RPM, i.e., the speed of the piston rotation within the cylinder, and volume strokes, i.e., how many times the piston rotates within the cylinder in the pump per trigger. Besides the RPM and volume stroke, the displacement on the pump can also be used to adjust the size of drops generated. Drop dynamics is also known to be strongly influenced by surface tension and viscosity and can be characterized by dimensionless quantities, such as the Reynolds (Re), Weber (We), and Ohnesorge (Oh) numbers, which are shown in eqs 1−3.19,20 Re =

νρd μ

We =

Oh =

ν 2ρd σ

μ = ρ dσ

(2)

We Re

(3)

where ρ = density [kg/m ], σ = surface tension [N/m], ν = velocity [m/s], μ = dynamic viscosity [kg/s m], and d = characteristic length (nozzle diameter) [m] The Weber number relates the fluid’s inertia to its surface tension. There is a minimum Weber number before which drop breakup does not occur because a drop must have sufficient energy to overcome the fluid/air surface tension at the nozzle exit.20 After the drop is deposited, it should leave a single isolated spread drop. The mechanisms that lead to splashing are still subject to research.21 The Ohnesorge number relates the viscous force relative to the surface tension force. At high values of Oh, viscous dissipation prevents the drop 3

(1) F

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX

Article

Industrial & Engineering Chemistry Research

volume strokes of 2.) For melt-based formulations containing polymers as the carrier, volume strokes 2 is the optimum operating region. However, volume strokes 1 and 3 are also tested, and good drop quality could not be achieved. The data for all volume strokes is saved in the OperabilityData node, not shown here. Next, the GraphData workflow is used to generate a visual operating region for each formulation. In Figure 11, for the same experiment with workflow ID 382, the operating region (drop quality = 1) is plotted as a combination of pump displacement and pump rate. With the help of this plot, the operator can chose from the pump displacement and rates to generate good drops at pump volume strokes 2. 5.2. Operating Regime Based on Dimensionless Numbers. In general it is desirable to characterize the regions of operating space in which feasible drop formation can take place in a generalizable form, which can then be used in predictive mode. Traditionally, this is accomplished through the use of dimensionless numbers such as Re, We, and Oh which include measured values of fluid parameters. By way of example, we consider two solvent-based formulations: one which consists of 30% naproxen and 70% PVP K90 and the other of 70% naproxen and 30% PVP K90 dissolved in ethanol. The measured properties of density, surface tension, and viscosity values, listed in Table 1, would be recorded as part of the information associated with the input material node to the workflow. The density is calculated by measuring the weight of solution per unit volume. The surface tension is measured with pendant drop method, and viscosity is measured with a rotating cylinder rheometer (Brookfield Engineering Laboratories, Inc., Massachusetts). The fluid velocity must also be measured experimentally. At each printing setting, the pump is primed for 30 s, and the volume of the printed drops is measured with a graduated cylinder. Three measurements are recorded for each setting, and the average value is used in the calculations. With the known nozzle diameter and cross sectional area, the fluid velocity can be calculated. With these characteristics determined and recorded, the Reynolds, Weber, and Ohnesorge numbers can be calculated using eqs 1−3 and the measured values of velocity, density, nozzle diameter, viscosity, and surface tension. (The drop formation quality information on solvent-based formulation 30NAP70PVP along with measured velocity and Re, We, and Oh number values are shown in Table S.2 of the Supporting Information.) An operating region graph using Weber and Reynolds numbers can be constructed as shown in Figure 12. Although a clear distinction between different regions cannot be made given the limited data, the black arrow demonstrates the transition from printable region to the region where tail and satellite drop formation is observed, followed by the region where there is insufficient energy for drop formation. The outlier points shown in Figure 12 might be due to experimental errors while measuring various parameters. Similar studies to determine the operating regime using the dimensionless numbers have been reported for Newtonian fluids.20 However, PVP is known as a viscoelasticity enhancer polymer, and thus, polymer solutions used in this study are non-Newtonian.24 In general the development of operating regions for various classes of non-Newtonian fluids is a timeconsuming activity. However, since KProMS allows systematic capture and retention of this level of experimental data, including the fluid properties and other factors needed to define

from ejecting, and at low values of Oh, the primary droplet is followed by multiple satellite droplets.22 Satellite drops are undesirable because they are detrimental to the precision of drop deposition.23 5.1. Operating Regime Based on Printing Parameters. For a selected drug formulation, the best operating conditions often have to be determined experimentally because firstprinciples predictive models are not available for the typical non-Newtonian fluids which arise in formulations. A combination of operating variable values is acceptable if it produces good quality drops where good quality drop formation constitutes reproducible formation of single primary drop with no tail, satellite drops, or spraying. Drop specification criteria are demonstrated in Figure 9 along with examples of on and

Figure 9. Drop specification 362 criteria.

off specification drops. The images are taken during operating regime determination of various melt-based formulations. The workflow, named TB3OpRegime, for determining the operating regime for a given formulation is shown in Figure 10. The data node named OperabilityData represents the operating

Figure 10. Workflow for determining the operating regime.

regime data created for one combination of nozzle and formulation. An instance of this workflow represents an experiment performed to determine the operating regime for one combination of formulation, nozzle size, and operating temperature. The test bed is operated at different settings for pump displacement, pump rate, and number of volume strokes. The data file associated with the data node has the following four columns: pump rate (RPM), pump displacement, number of volume strokes, drop quality. The first three values in each row in the file represent one set of operating conditions, and the last value represents the quality of drops generated (0 for bad drops, 1 for good drops) under those operating conditions. The quality determination can be made on the basis of operator observation of drop features, as shown in Figure 9 or through automated image analysis. The operating regime data corresponding to the experiment with workflow ID 382 can be found in Table S.1 of the Supporting Information. The experiment is performed with a melt-based formulation consisting of 85% PEG3350 and 15% naproxen, and with reservoir, tubing, and pump temperatures controlled at 87.5, 90, and 140 °F, respectively. The nozzle N19 is used which has an inner diameter of 0.912 mm (19 AWG). (In Table S.1, operating regime data is shown only for pump G

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX

Article

Industrial & Engineering Chemistry Research

Figure 11. Operating region plot.

formulations are introduced. The resulting sets of printability regions represent knowledge that has predictive value for guiding the selection of conditions for new formulations. The structured KProMS workflow framework allows operating data to be accumulated in a systematic way so as to facilitate generalization from the multiple instances collected, which is one form of knowledge creation.

Table 1. Fluid Properties for Solvent-Based Formulations density (g/mL) interfacial surface tension (mN/m) viscosity (mPa s)

30NAP70PVP

70NAP30PVP

0.83 24.48 73.90

0.83 24.14 16.37

6. ADDITIONAL KPROMS FUNCTIONALITIES As demonstrated in the previous sections, KProMS provides the complete context of the data entered and generated during the execution of the associated workflow, and the metadata of all data associated with the data nodes in a workflow. Another key aspect of data provenance that is important in the pharmaceutical industry is providing assurance of data accuracy. This is facilitated in KProMS through access privileges managed by the underlying HUBzero framework. The access to the use of KProMS can be controlled through user login, groups, and authority levels. For example, the person assigned as the “supervisor” of a workflow plays several key roles. Only the supervisor can approve and submit an instance of a workflow. Prior to submitting an instance, the supervisor can check the plausibility and acceptability of the associated data. Additionally, once an instance is saved in KProMS, the built-in safeguards of the MySQL database system make it virtually impossible to destroy or tamper with the data. Of course, KProMS is not as yet commercialized and therefore has not gone through a formal validation process. However, to the extent that the HUBzero framework allows, accuracy of data is

Figure 12. Operating regime graph for solvent formulation naproxenPVP K90 (30:70).

the relevant dimensionless numbers, a generalizable printability region for specific rheological classes of fluids can be developed over time from workflow instances accumulated as new H

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX

Article

Industrial & Engineering Chemistry Research

facilitates root cause determination if deviations are found to arise subsequently. (8) Structured data facilitates extraction of relevant data to support predictions quantitative or qualitative of conditions for new instances. (9) Hierarchical structure allows recording of Supporting Information and analyses such as DOE design, off-line laboratory analyses, calibration sets, etc., by means of supporting workflows, which are hierarchically linked to the primary workflow.

assured through accountability, and trustworthiness is assured by the database functionalities. The 10 aspects of highly effective research data25 are Stored, Preserved, Accessible, Discoverable, Citable, Comprehensible, Reviewed, Reproducible, Reusable, and Integrated. The workflow-based framework of KProMS provides the necessary functionalities to impart these aspects to the data managed by it. Thus, the associations between different aspects of data form the base for knowledge generation. The ability to represent the relationships between data through a hierarchy of workflows is another important feature of KProMS. The various workflows discussed earlier about specific experiments and calculations are in fact related through a hierarchy of workflow, which represents an overall project management function.3 For example, the overall objective may be to generate operating regimes based on Reynolds and Weber numbers. A workflow for achieving that objective is shown in Figure 13. When the experimental procedures explained in the paper are applied to a wide variety of drug formulations spanning a range of properties, the Operating Regime based on corresponding properties and operating conditions becomes the single source for checking and/or ensuring drop quality for specific operating conditions such as volumetric flow rate. Thus, the workflow shown in Figure 13 represents the relationships between various experiments and computations for

7. CONCLUSIONS A knowledge management system is an essential tool for recording and making accessible the data provenance of each and every dosage produced via DAMPP. The processing steps in the DAMPP system were modeled in the workflow-based Knowledge Provenance Management System, KProMS. The integration of the automation program and KProMS allows inputting all process settings data only once through KProMS and successfully running the automation program to execute the process operation. The ability to save the output data file, drop images, IR images, and crystallinity data after each process run allows easy access to all process-related online or offline generated data. KProMS is readily and effectively used for managing the data associated with the experimental studies. Through statistical analysis of production runs, the experiments can be compared and the best process conditions determined for the desired outcome, i.e., dosage amount. Using KProMS, operating regions based on printing parameters can be generated for various formulations, and this knowledge can be used subsequently to select the best operating conditions for executing the process workflow with a new formulation. While such operating regions can be generated through a DOE-based study for each new formulation, the operating region is preferably defined in terms of a printability region based on dimensionless numbers for classes of fluid rheologies. In order to develop such a printability region accurately, systematic experimental data needs to be accumulated over a number of formulations, with fluid parameters determined for each formulation. KProMS provides the structured framework to archive data accumulated over an extended period time from successive studies with specific formulations to develop and update such a printability region as guidance for choosing operating regions for future applications with different formulations. Building this form of knowledge is greatly facilitated by a structured KM system.

Figure 13. Workflow for comprehensive operating regime determination.

each drug. Each activity or experiment, in general, will invoke a workflow, which represents the specifics of that step. For example, the TB3Basic step will implement the TB3Basic workflow shown in Figure 2 and so on. The data in the Operating Regime step will be a compendium of the data generated for individual drugs. A summary of the key functionalities provided by KProMS, not all of which could be illustrated with the example provided in this paper, is given as follows: (1) There is a graphical representation of procedures to be followed so as to build consensus on how the activities should be carried out and share understanding of what will be done. (2) Definition of metadata is associated with parameters and data recorded (names, units, etc.). (3) As an instance of the workflow being executed, the workflow provides the framework for recording parameters and storing results or data output. (4) The workflow structure enforces the collection of required information, reducing occurrence of data gaps because such gaps are visible. A workflow control feature can be invoked which does not allow entry of data associated with a downstream task until all of the data associated with the upstream task has been recorded. (5) Workflow progression provides the status of the information generated to that point both in terms of completion of an instance and completion of a set of instances. All data is transparent and accessible to those provided viewing privileges. (6) Set of workflows becomes a knowledge source to guide selection of conditions for other instances. (7) Structured data



APPENDIX Figure A.1 shows the workflow symbols. Explanations of data entry form generation which are shown in Figure 2 of the main text and Figure A.2 follow: (a) Through the blue TB3Basic workflow icon shown in Figure 2 of the main text, the data entry form can be accessed. This allows the user to

Figure A.1. Workflow symbols. I

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX

Article

Industrial & Engineering Chemistry Research

Figure A.2. (a) Data entry form for Material Source R1. (b) Data entry form for IVEK pump setup task.

Figure A.3. Input file in CSV format.

Figure A.4. Output file in CSV format.

LabviewData, IRData, and CrystData represent the data created by Labview, IR camera image files, and crystallinity data for the dosage forms, respectively. Example files are shown in Figures A.3 and A.4.

specify necessary information about the run, and it is shown in Figure 3. (b) The material sources (yellow pentagon: R1 and Subs) represent the formulation and substrate used. The data entry form for formulation material source R1 is shown in Figure A.2a. A yellow triangle represents the material produced. (c) The workflow consists of the five tasks (green rectangles: Reservoir, IVEK, Nozzle, Staging, and Cryst) described previously. The data entry form for IVEK pump setup task is shown in Figure A.2b. (d) A camera captures the image of each drop released from the nozzle, and an IR camera captures the image of drops solidifying on the substrate. The data nodes named



ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.iecr.6b01042. Operating regime data for PEG:NAP (85:15) with nozzle N19 and workflow ID 382 and dimensionless number J

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX

Article

Industrial & Engineering Chemistry Research



(14) Muñoz, E.; Capón-García, E.; Laínez-Aguirre, J. M.; Espuña, A.; Puigjaner, L. Using mathematical knowledge management to support integrated decision-making in the enterprise. Comput. Chem. Eng. 2014, 66, 139−150. (15) Quaglia, A.; Sarup, B.; Sin, G.; Gani, R. Integrated business and engineering framework for synthesis and design of enterprise-wide processing networks. Comput. Chem. Eng. 2012, 38, 213−223. (16) Abbott Informatics. Electronic Laboratory Notebook (ELN); https://www.abbottinformatics.com/us/products/eln (accessed Mar 5, 2016). (17) Dassault Systemes. BIOVA LIMS; http://accelrys.com/ products/pdf/pmc/biovia-lims-ds.pdf (accessed Mar 5, 2016). (18) McLennan, M.; Kennell, R. HUBzero: A Platform for Dissemination and Collaboration in Computational Science and Engineering. Comput. Sci. Eng. 2010, 12 (2), 48−53. (19) de Gans, B.-J.; Duineveld, P. C.; Schubert, U. S. Inkjet printing of polymers: state of the art and future development. Adv. Mater. 2004, 16 (3), 203−213. (20) Derby, B. Inkjet Printing of Functional and Structural Materials: Fluid Property Requirements, Feature Stability, and Resolution. Annu. Rev. Mater. Res. 2010, 40 (1), 395−414. (21) Derby, N.; Reis, B. Inkjet printing of highly loaded particulate suspensions. MRS Bull. 2003, 28, 815−818. (22) Reis, N.; Derby, B. Ink Jet Deposition of Ceramic Suspensions: Modeling and Experiments of Droplet Formation. MRS Online Proc. Libr. 2000, 624, 65−70. (23) Dong, H. M.; Carr, W. W.; Morris, J. F. An experimental study of drop-on-demand drop formation. Phys. Fluids 2006, 18 (7), 072102. (24) Tuladhar, T. R.; Mackley, M. R. Filament stretching rheometry and break-up behaviour of low viscosity polymer solutions and inkjet fluids. J. Non-Newtonian Fluid Mech. 2008, 148 (1−3), 97−108. (25) de Waards, A.; Cousijn, H.; Aalbersberg, I. 10 Aspects of Highly Effective Research Data; https://www.elsevier.com/connect/10aspects-of-highly-effective-research-data (accessed Jul 16, 2016).

calculations for solvent-based formulation naproxenPVPK90 (30:70) (PDF)

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Notes

The authors declare no competing financial interest. Author Contributions †

K.L. and J.S. are equal contributors.



ACKNOWLEDGMENTS This work was funded by the National Science Foundation under Grant EEC-0540855 through the Engineering Research Center for Structured Organic Particulate Systems. The authors would like to thank Christopher Anthony for his help with measuring surface tension and viscosity of selected formulations.



REFERENCES

(1) Içten, E.; Giridhar, A.; Taylor, L. S.; Nagy, Z. K.; Reklaitis, G. V. Dropwise Additive Manufacturing of Pharmaceutical Products for Melt-Based Dosage Forms. J. Pharm. Sci. 2015, 104 (5), 1641−1649. (2) Hirshfield, L.; Giridhar, A.; Taylor, L. S.; Harris, M. T.; Reklaitis, G. V. Dropwise additive manufacturing of pharmaceutical products for solvent-based dosage forms. J. Pharm. Sci. 2014, 103 (2), 496−506. (3) Joglekar, G. S.; Giridhar, A.; Reklaitis, G. A workflow modeling system for capturing data provenance. Comput. Chem. Eng. 2014, 67, 148−158. (4) Içten, E.; Joglekar, G. S.; Giridhar, A.; Reklaitis, G. V. Application of a Knowledge Management System to a Dropwise Additive Manufacturing System for Pharmaceuticals. Comput.-Aided Chem. Eng. 2016, 38, 619−624. (5) IVEK Corporation. Single Channel Dispensing Systems; http:// www.ivek.com/SC_Dispensing_Systems.html (accessed Feb 26, 2016). (6) Hirshfield, L.; Içten, E.; Giridhar, A.; Nagy, Z. K.; Reklaitis, G. V. Real-Time Process Management Strategy for Dropwise Additive Manufacturing of Pharmaceutical Products. J. Pharm. Innov. 2015, 10 (2), 140−155. (7) Içten, E.; Nagy, Z. K.; Reklaitis, G. V. Process control of a dropwise additive manufacturing system for pharmaceuticals using polynomial chaos expansion based surrogate model. Comput. Chem. Eng. 2015, 83, 221−231. (8) Içten, E.; Giridhar, A.; Nagy, Z. K.; Reklaitis, G. V. Drop-onDemand System for Manufacturing of Melt-based Solid Oral Dosage: Effect of Critical Process Parameters on Product Quality. AAPS PharmSciTech 2016, 17 (2), 284−293. (9) Bayer, B.; Marquardt, W. Comparison of Data Models in Chemical Engineering. Concurr. Eng. Res. Appl. 2003, 11 (2), 129−138. (10) Gabbar, H. A.; Aoyama, A.; Naka, Y. Automated solution for control recipe generation of chemical batch plants. Comput. Chem. Eng. 2005, 29 (5), 949−964. (11) Wang, B. G.; Wright, A. R.; Morris, A. J. A prototype recipe management system in batch processes. Comput. Chem. Eng. 1997, 21 (11), 1311−1323. (12) Venkatasubramanian, V.; Zhao, C.; Joglekar, G.; Jain, A.; Hailemariam, L.; Suresh, P.; Akkisetty, P.; Morris, K.; Reklaitis, G. V. Ontological informatics infrastructure for pharmaceutical product development and manufacturing. Comput. Chem. Eng. 2006, 30 (10− 12), 1482−1496. (13) Suresh, P.; Hsu, S.; Akkisetty, P.; Reklaitis, G. V.; Venkatasubramanian, V. OntoMODEL: Ontological Mathematical Modeling Knowledge Management in Pharmaceutical Product Development, 1: Conceptual Framework. Ind. Eng. Chem. Res. 2010, 49 (17), 7758−7767. K

DOI: 10.1021/acs.iecr.6b01042 Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX