37.2.1 Development and Scale‐Up in Agitated Filter Dryers

The API compound is a channel solvate crystallized from a dual solvent solution. The crystal slurry is filtered to isolate the compound. The filter cake resistance and compressibility as measured by a leaf filter were low, indicating good pressure filtration (and centrifugation) characteristics. The crystal slurry filtration and cake wash are followed by a two hour nitrogen blow to deliquor and partially dry the wet cake. This reduces the wet cake solvent content to approximately 8–12 wt %.

TGA measurements of wet cake with an initial solvent content of 27 wt % showed that the drying rate starts falling when the solvent content drops to 11–12 wt % at which point the remaining solvent is largely internal to the crystals. The solvent content of the fully solvated crystal is 19 wt %. Based on TGA measurements, a minimum temperature of 80 °C is required to achieve a reasonable desolvation rate; these results were confirmed by means of drying rate measurements performed in a lab vacuum tray dryer at several temperatures. Based on these results, the final drying temperature was increased from 50 to 80 °C. The jacket temperature is ramped from 20 to 80 °C over six hours to limit the rate of solvent vapor generation, avoiding condensation in the dryer and vacuum line.

The propensity for particle attrition and agglomeration was measured in a lab‐scale AFD over a solvent content range of 28–0 wt %. There was no agglomeration or snowballing; however, attrition occurred over the entire range and was particularly high at high solvent concentrations. In scaling up to the pilot plant and production AFDs, attrition was managed by reducing the solvent content (by nitrogen cake blow) prior to initiating AFD blade rotation and by minimizing the number of AFD blade rotations. Agitation was initiated at low cake solvent content in which the particles are less sensitive to stress. The AFD blade is typically rotated 40 times at the lowest speed every two hours, achieving a total of approximately 900 revolutions during drying. Attrition was also measured in pilot‐scale AFDs. One batch that was exposed to a large amount of agitation (2200 revolutions) exhibited a approximately 50% reduction in particle size (see Table 37.1); this was greater attrition than a similar batch agitated for 610 revolutions at a slightly lower tip speed.

37.2.2 Production‐Scale Equipment Change

After successfully completing many batches at the production site, a switch from an AFD to a centrifuge and CSD was evaluated in order to increase throughput and free up the AFD equipment train for another compound. The modified process was developed at lab and pilot scales and then scaled up and demonstrated at production scale.

37.2.3 Switch from Pressure to Centrifugal Filtration

Centrifugation was demonstrated using a lab‐scale 12 cm diameter vertical bowl centrifuge with perforated bowl (Figure 37.1). This centrifuge is capable of generating a g‐force of approximately 600 g, which is less than 1000–1500 g at large scale. An additional limitation of the lab‐scale centrifuge is the low liquid head pressure (~15 psi) that is generated; this is less than the 100–200 psi at large scale. The decision to proceed to pilot scale was supported by polarized light microscopy (PLM) and particle size analyses that showed no significant fracture of crystals in centrifuged wet cake and wet cake from high pressure filtrations.

Centrifugation was evaluated using two pilot‐scale centrifuges. The solvent content of washed centrifuge wet cake (~20 wt %) is higher than that of deliquored AFD wet cake (~8–12%) since the two hour nitrogen blow of the AFD wet cake is effective in evaporating a large fraction of the more volatile solvent (dichloromethane (DCM)), but less effective in removing the higher boiling ethyl acetate (EtAc) antisolvent. Consequently, the deliquored AFD wet cake contains a higher concentration of EtAc internal to the channel solvate crystals than centrifuge cake. In order to achieve similar solid‐state drying behavior in the two different equipment trains, similar wet cake solvent compositions prior to the start of heated drying were targeted. One approach to accomplish this is to sweep nitrogen through the washed wet cake in the centrifuge bowl; however, the available production‐scale centrifuge lacked this capability. An alternative approach consisting of modifying the composition of the centrifuge cake wash solution was pursued.

The primary purpose of the cake wash is to remove the mother liquor containing dissolved impurities from the cake. In order to ensure effective impurity removal, a plan to apply two sequential washes to the centrifuge cake was considered. The composition of the first wash was the standard 0.6 : 1 (w/w) EtAc: DCM; this composition approximates the mother liquor composition. This wash was followed by a wash solution with high EtAc composition. EtAc is an antisolvent, so the second wash may be less effective in removing impurities. This approach was rejected because it required dual solvent wash tanks for multiple centrifuge loads; however, just a single wash tank was available at the production site.

A single wash with high EtAc content was evaluated in lab‐scale centrifugation experiments (see Table 37.2). EtAc contents from 1.3 : 1 to 6.7 : 1 (w/w) EtAc : DCM were tested. Application of a wash of 2.7 : 1 EtAc : DCM increased the composition of the residual solvent in the centrifuge cake to 1.5 : 1; a further increase in the EtAc content of the applied wash did not further increase the EtAc content in the wet cake. The wash amount (6.4 kg/kg dry solids) used for the centrifugation matched the total wash amount used for the pressure filtration in the AFD. However, the wash solution has a short contact time with the centrifuge wet cake; this provides limited opportunity for DCM in isolated pockets of the wet cake to diffuse to the main flow channels in the cake. DCM present in the internal channels of the solvate crystals is more difficult to remove by the cake wash solution. The single wash with high EtAc content was shown to achieve acceptable removal of impurities present in the mother liquor, and the wash ratio of 2.7 : 1 was implemented.

X‐ray powder diffraction (XRPD) analyses of lab‐dried wet cake showed that a residual solvent EtAc–DCM ratio of 1.0 : 1 (w/w) or greater at the start of heated drying generates the desired solid structure when the wet cake is dried under appropriate conditions. So a residual solvent composition of 1.0 : 1 or greater was targeted for the centrifuge wet cake.

Residual solvent compositions exceeding 1.0 : 1 EtAc : DCM (w/w) were achieved in spun‐down centrifuge wet cake in pilot plant batches using a wash solution composition of 2.7 : 1 EtAc : DCM (Table 37.3) for vertical bowl and horizontal peeler centrifuges (Figure 37.2).

Operating conditions and process data for representative centrifugations are summarized in Table 37.4. The g‐forces used for slurry loading/cake washing and for final spin down in the pilot batches with the largest pilot plant centrifuge (horizontal peeler) were similar to those implemented at production scale.

The wet cake produced using the pilot‐scale centrifuge with highest g‐force exhibited an acceptable degree of attrition relative to the crystal slurry feed (Table 37.5).

The process was transferred to the production‐scale train, which employs a horizontal inverting bag centrifuge. Application of a cake wash with a 2.7 : 1 (w/w) ratio of EtAc : DCM generated a final wet cake residual solvent composition of 1.1 : 1 EtAC : DCM; this met the target of greater than or equal to 1.0 : 1. In the new production train, the centrifuge wet cake is transferred to a CSD for the subsequent drying operation.

37.2.4 Development and Scale‐Up in Conical Screw Dryer

Advantages of CSDs include high heat and mass transfer coefficients resulting from agitation that exposes fresh wet solids to the heated surfaces and headspace. Screw rotation achieves mixing in the radial and axial directions, while screw orbiting mixes in the angular dimension. The gentle agitation introduces relatively little stress, resulting in less particle attrition for many compounds.

Lab‐ and pilot‐scale drying studies were performed to develop and scale up the conical screw drying to a 3000 L production‐scale dryer. CSD equipment information is summarized in Table 37.6. Large CSDs have significantly less heat transfer area per volume of cake than a lab‐scale CSD. All of the dryers have a tapered screw that achieves better mixing and conveying of solids near the top of the dryer than a cylindrical screw. The screws for the lab‐ and pilot‐scale CSDs have closed flights that provide more area for vertical solid conveyance, while the production‐scale dryer screw has open flights (lacking recesses) that facilitate screw cleaning.

Since the drying rate of the lab 1.5 L CSD (Figure 37.3) is not readily scalable to large dryers, the lab experiments focused on the impact of drying process parameters and mechanical stress on particle physical properties (e.g. attrition and agglomeration). In one experiment, drying with constant agitation (at 40% of maximum speed) was continued for 17 hours after drying completion; this extended period of agitation of the dried particles resulted in essentially no particle attrition. Similarly, agitation at maximum tip speed did not attrite the particles. The ease of heel removal from the wall during dryer unloading was also assessed. A mass spectrometer was employed to monitor drying rate, facilitate the decision on drying completion, and help detect an improper mechanical seal in the dryer. Wet cake from the pilot‐scale centrifuge was used for these studies.

Pilot‐scale studies in the 30 L CSD using centrifuge wet cake were performed to measure the effect of different profiles of agitation, vacuum pressure, and jacket temperature on drying rate and dried particle properties. Heat and mass transfer rates of the 30 L dryer can be used for scale‐up (within certain constraints); this is typically done by measuring the drying rate of a pilot‐scale batch employing the design temperature, pressure, and agitation profiles. The drying rate is expressed as a function of the residual solvent content (see Figure 37.4). The resulting drying rate curve is typically divided into three or more regions that correspond to different rate‐controlling mechanisms.

The drying rate for a particular region of the curve is similar across scales. The drying time for each region is calculated separately using Eqs. (37.1) and (37.2):

(37.1)

(37.2)

where

• (hours).
• is the falling rate region 1 drying time (hours).
• m_dry is the dry mass (kg).
• X is the solvent content of cake (kg solvent/kg dry cake).
• A is the heat transfer area (m²).
• ṁ_c is the drying rate of constant rate region (kg/m²/h).

The total drying time is the sum of the individual drying times: t_total = t_C + + . Scale‐up using the drying rate curve approach assumes equivalency across scales of heat and mass transfer coefficients, temperature driving forces, vacuum pressure, nitrogen sweep, mixing, dryer geometry, and cake characteristics.

The mixing turnover time is calculated using equations 37.3 and 37.4 and is maintained across scales:

(37.3)

(37.4)

where

• D is the screw outer diameter (cm).
• d is the screw shaft diameter (cm).
• n is the screw rotation speed (rpm).
• P is the screw pitch (cm).

For open flight screws, the screw shaft diameter is replaced with an equivalent diameter selected to reflect the reduced flight area available for conveyance. For the case at hand, a mixing turnover time of three minutes is achieved in the pilot‐ and production‐scale CSDs (filled to capacity) using screw rotation speeds of 11 and 45 rpm, respectively. The ratio of screw rotation speed (rpm) to orbiting speed (rpm) is maintained at 50–100 to ensure mixing in all three spatial dimensions.

Solvent vapor is removed through a vacuum port on the top of the CSD. The upward transport of solids by the screw brings fresh solids to the surface, resulting in a high vapor mass transfer coefficient. Nitrogen may be applied to the bottom of the CSD to further improve mass transfer by providing a convective sweep resulting from the flow of N₂ from the bottom to the top of the bed. In pilot studies, it was found that substantial particle attrition resulted from a high N₂ gas velocity; similarly high attrition also resulted from a leak in the bottom discharge valve of the CSD that led to airflow into the bottom of the evacuated CSD. For this reason, N₂ was not applied to the bottom of the CSD, and the importance of a pre‐batch vacuum leak test of the CSD was emphasized. For other less stress‐sensitive products, application of a slow N₂ flow to the CSD bottom may be appropriate; in this case, the additional flow area provided by a sintered N₂ distribution ring at the bottom of the CSD may result in reduced N₂ gas velocity and less particle attrition in comparison with an N₂ distributor consisting of just a few drilled holes that provide a small N₂ injection area.

A low level of attrition of crystals during agitated drying was measured in lab‐ and pilot‐scale CSD tests; consistent results were obtained in the plant CSD. Table 37.7 compares the size of the crystals composing the centrifuge wet cake charged to the CSD with the size of the dried crystals. The degree of attrition in the CSD is less than half of the attrition measured in the AFD – despite the use of continuous agitation in the CSD. For both dryer types, agitation was initiated soon after the dryer jacket reached its final temperature of 80 °C.

The information generated by the lab‐ and pilot‐scale studies enabled a successful scale‐up and demonstration batch in the production‐scale CSD. The desired crystal structure was achieved by desolvating the crystal solvate at high temperature by rapidly heating the dryer jacket to 80 °C with no agitation and limited vacuum. The vacuum level in the initial drying phase was sufficient to avoid vapor condensation in the CSD and vacuum exhaust line; it also limited the amount of vaporization and resulting cooling. Agitation and high vacuum were applied after the jacket temperature reached 80 °C; the agitation distributed the heat imparted by the jacket. Residual solvents were reduced to passing levels within 36 hours of drying after high vacuum was achieved.

37.3.1 Desolvation Kinetics and Form/Thermal Stability

The intrinsic desolvation kinetics were calculated by measuring the solvent (ethanol) content as a function of time in dynamic vapor sorption (DVS) experiments carried out at three different temperatures (40, 60, and 70 °C) under nitrogen sweep. A thin layer of solids and high N₂ sweep rate were used to minimize the mass transfer resistance and ensure measurement of intrinsic desolvation rate. The net desolvation rate constant (k_D; units of s⁻¹) for each temperature was calculated by fitting the desolvation data to first‐order kinetics. The resulting rate constants were graphed in an Arrhenius plot to quantify the temperature (T in K) dependence:

(37.5)

Drying results of early pilot plant batches were consistent with the findings of the desolvation kinetics study. In early pilot plant batches, the dryer was operated at a nominal jacket temperature of 50 °C, resulting in a maximum internal temperature of about 40 °C; the drying process took nearly a week at this temperature to reduce the ethanol content to an acceptable level (≤0.1 wt %).

Thermal stability and form stability studies were performed to determine the maximum acceptable drying temperature. At temperatures above 80 °C, wet cakes began to discolor – indicating thermal degradation. For the case of properly deliquored wet cake, the desolvate was shown by XRPD to be the stable crystal form at temperatures below 80 °C, which is well below the API melting temperature of approximately 150 °C. Amorphization of the desolvate crystal was not observed under these conditions.

37.3.2 Composite Drying Model

Drying of this compound consists of two main phases. The first phase is relatively short and consists of rapid evaporation of unbound solvent accomplished by the nitrogen cake blow that follows proper cake deliquoring. Desolvation is insignificant during this phase due to its high energy barrier.

The second phase consists of removal of bound solvent (primarily as solvent internal to the crystals) in three steps:

1. Desolvation: The first step is the desolvation of the ethanol solvate. The net desolvation rate (R_D) is governed by temperature. The desolvation results in the separation of solvent in liquid form:
   
2. Evaporation: The second step is the evaporation of the free liquid solvent after desolvation:
   
3. Convection: In the third step, the evaporated solvent is convected away by the nitrogen sweep. The nitrogen is expected to flow around the particles at a sufficiently high velocity to alleviate diffusion limitations in this third step.

The desolvation and evaporation rates depend on the local temperature and heat transfer rate, which are governed by the physical properties of the cake, jacket fluid temperature, heat transfer area, and the mixing provided by the agitator. By applying a sufficient nitrogen gas flow rate for rapid convection of solvent vapor, desolvation or evaporation becomes the rate‐determining step. The effective rate of solvent loss is the smaller of the desolvation rate and the solvent evaporation rate. At a given temperature T, the desolvation rate is readily calculated using the equation R_D = k_D(T)C_s,b, where C_s,b is the ratio of the mass of the bound solvent in the form of solvate to the mass of desolvated solids.

The heat transfer model used in the simulation is based on the penetration theory proposed by Tsotsas and Schlunder [1] with a few modifications. The heat transfer mechanism is dominated by thermal conduction. The AFD is divided into three regions in the heat transfer analysis: (i) the vessel wall, (ii) the gap between the outer layer of the dry solid bed and the vessel wall (this is assumed to be composed of the first layer of solids and gas gap), and (iii) the dry solids. Further description of the mathematical modeling is provided in Ref. [2].

In the absence of agitation during drying, a gas gap is created by preferential flow of N₂ along the vessel wall, generating additional resistance to heat transfer. The contribution of this gas gap to the overall heat transfer resistance was found to be 10–50% depending on the fraction of wet solids; this justifies its incorporation in the heat transfer model. The model also incorporates an equation for the contact heat transfer coefficient (h_ws) by Hoekstra et al. [3] for the first layer of solids adjacent to the dryer wall; h_ws is inversely dependent on the mean free path length for N₂ gas molecules. Use of this equation enables simulation of the effect of dynamic pressure in the filter dryer. The pressure variation may be significant due to the N₂ gas flow rate, the vacuum drawn at the AFD bottom, and dynamics of pressure drop across the cake. A headspace pressure increase from 50 to 1000 mbar absolute increases h_ws by 2.5x.

37.3.3 Regression of Model Parameters Through Simulations

All simulations were carried out using the modified model framework within a commercially available simulation software. The contact time for mixing was calculated from the mixing number, N_mix, which is empirically related to the Froude number. Two constants in the N_mix correlation that are characteristic of the agitator type were estimated by forcing a reasonable fit of the model results with the temporal profiles of the measured solvent content and solids temperature of an early pilot plant batch (Figures 37.5 and 37.6).

The initial solvent composition of 5.4 wt % in Figure 37.5 represents the first available measurement. This solvent content is below the bis‐solvate solvent composition of 8.8 wt %, indicating that the crystals are partially desolvated when heated drying is initiated after the completion of the nitrogen blow of the cake. Note in Figure 37.6 the initial phase where the solids temperature increases (from ~19 °C used for the N₂ blow) before attaining a somewhat steady value. The time at which the internal temperature achieves a somewhat steady value corresponds to mono‐solvate composition based on Figures 37.5 and 37.6. The first solvent molecule of the bis‐solvate likely has a lower energy barrier for desolvation than the second solvent molecule. During removal of the first solvent molecule, evaporation of the desolvated solvent may be the rate‐determining step, thereby increasing the solids temperature. The thermocouple tip is flush with the AFD wall surface and is in contact with the gas flowing through the gap between the dryer wall and the solids. Consequently, the measured temperature does not exactly match the dry solids temperature predicted by the model. This is supported by the observed jumps in the measured temperature during the mixing period as shown in Figure 37.6.

37.3.4 Improved Drying Through Model Simulations

Simulations were performed using the computational model to better understand the dynamics of agitated drying. The heat transfer rate is dependent on the jacket temperature and the agitation speed and time. A jacket temperature upper limit of 70 °C was imposed to assure thermal stability of the product and acceptable crystal form, which in this case demanded a gradual increase in the drying temperature. It was seen that the agitation frequency, duration, and speed play a significant role in enhancing the drying rate. Continuous agitation results in the fastest drying at a given jacket temperature. Since agitation can also result in significant attrition of product crystals, intermittent agitation is commonly used in practice. The mixing period and intensity that produce acceptable particle attrition form a practical basis for the design of the agitator operating protocol. Model simulations were run using various headspace pressures and agitation profiles and yielded the improved drying protocol in Figure 37.7.

Prior to the start of heated contact drying, the wet cake is blown with nitrogen for three hours. The jacket temperature is then gradually increased over 12 hours to avoid excessive heating of wet cake with unbound solvent as it may result in the formation of a dry coating on the jacketed wall, reducing the heat transfer rate. No agitation is performed at high cake solvent content to avoid agglomeration induced by mixing that can generate hard lumps upon heating – potentially reducing the drying rate. The original and improved drying protocols are compared in Table 37.8.

In the improved drying protocol, the agitation speed is reduced to reduce particle breakage and attrition, while the total mixing time is increased. This protocol was implemented in a pilot plant batch. Figure 37.8 compares the model predictions and observed drying performance, demonstrating that the model simulations are in close agreement with the measured solvent content throughout the drying process.

The drying time to reduce the residual solvent content to 0.1 wt % (from 8.8 wt %) was approximately 35 hours. This constitutes a fourfold reduction from the 140 hours required in the early pilot plant batch. Four additional API pilot plant batches were dried using the improved operating conditions. Figure 37.9 depicts the temporal profiles of measured solvent content for all five batches dried using the improved protocol. All batches showed similar drying performance, thereby verifying the computational model description of the drying process. The model could not be verified beyond one set of operating parameters at scale due to constraints on the use of pilot plant equipment and the high API cost.

37.3.5 Guidelines for AFD Operation

On the basis of the model studies, pilot‐scale demonstration, and practical considerations, the following guidelines are recommended to achieve efficient AFD operation. In practice, the drying protocol will need to be tailored to address the specific issues related to the physicochemical characteristics of the particular product:

1. Efficient N₂ flow through the entire cake is important; the headspace pressure should not be allowed to increase beyond the point that causes cake compression, decreased inert gas flow rate, or increased mass transfer resistance. Intermittent cake smoothing must be performed to prevent preferential inert gas channeling or bypassing. This will alleviate mass transfer limitations relating to the movement of solvent vapors.
2. Cake that is substantially wet should not be mixed in order to avoid granulation. The wet granules can form hard lumps upon heating that can be difficult to dry, resulting in increased drying time.
3. Heat the wet cake gradually after ensuring deliquoring to the fullest practical extent. Heating wet cake with unbound solvent in the proximity of vessel wall may result in formation of a dry coat on the jacketed wall, causing fouling and a reduced heat transfer rate. An excessively high vaporization rate can also result in condensation within the dryer, resulting in formation of liquid droplets that may promote dissolution or granulation.