As far as the EDs are concerned, to obtain a meaningful assessment of battery lifetime and transmission range, two distinct devices were considered:
The main characteristics of the aforementioned devices are summarized in
Table 2. These two EDs have been chosen on a preliminary basis as potential candidates for the ADMIN 4D project, due to their flexibility, usability and performance, notwithstanding that they are characterized by a significant difference from a communication perspective, that is, the potential availability of an external antenna. Indeed, the Tinovi module features a dust and waterproof IP68 design, that does not allow the use of external antennas, forcing the use of the built-in antenna. Conversely, the Microchip mote is a node designed for showcasing increased degrees of freedom for testing and development in a broader set of applications. Although unable to provide any protection from water/dust, it supplies improved interfacing, and allows the connection of different kinds of external antennas. Clearly, different devices could have been adopted for the experimental strategy. For example, several types of LoRa EDs, equipped with sensor types, antennas and batteries (potentially) suitable for the ADMIN 4D project, are available off-the-shelf. However, the activity carried out was not meant as a comparison among device types but, rather, as a feasibility study of the proposed IoT based measurement system. For this reason, the assessment was performed on a limited number of devices. Based on the findings of this study, future development will include more comprehensive analyses involving several different products.
Using the hardware described, an extensive experimental campaign was carried out aimed at assessing the suitability of the chosen EDs for the project, particularly in terms of (i) guaranteed coverage range, and (ii) expected battery lifetime. A description of the tests undertaken and discussion of the relevant results are provided in the following paragraphs.
5.1. Coverage Range Tests
The goal of this set of tests was to assess the coverage range of the chosen sensors considering both the phases that characterize the application, production and final deployment, which highlight different issues from a communication point of view.
In the first test, an external Taoglas TI.08.C.0112 antenna [
32] was employed for the Microchip sensor, whereas the built-in antenna was used for the Tinovi. The goal was to investigate the behavior of the transmission even in highly critical scenarios, typical of the production phase, which can be represented by an indoor mixed industrial/office settlement where several electronic devices (e.g., smartphones, computers, 3D printers, etc.) are present, together with multiple walls, metallic structures and objects that may occlude the line of sight. A very good candidate for these tests was represented by one of the research building laboratories, located at the Department of Information Engineering, University of Padova (Italy), that hosts several laboratories, offices, and industrial machines over several floors, and hence meets the previous conditions. During the tests, the operational conditions of the production phase were emulated as much as possible. For this reason, the EDs were inserted in a box containing the same material used for the artifacts (river sands in this case), even if they were not progressively embedded, as occurs during the actual production phase. The LoRa GW and NS were placed inside the laboratory so as to have direct control of them, while the EDs were moved to several different locations within the Department at different distances and floors.
The experiments were performed with both the EDs, and adopting the same spreading factor,
. For a transmission test to be considered as passed, the LoRa gateway had to receive at least three-quarters of the transmitted data without errors, a proportion that has been considered to be able to ensure sufficient measurement continuity in the context of the ADMIN 4D project. The obtained results are reported in
Table 3, where “Yes” and “No” account for test passed or not, respectively.
The results showed that, on the one hand, LoRaWAN transmissions were quite reliable even in harsh environments, confirming the results obtained in other independent assessments [
33], where robustness of the network over the 868 MHz band for indoor operations was highlighted. On the other hand, the experiments showed, as expected, the more limited coverage range of the Tinovi sensor. Indeed, this sensor was unable to reliably receive packets beyond a range of 10–15 m, whereas the Microchip mote performed reliably within a 70 m range in the indoor experiments.
Nevertheless, it must be highlighted that both sensors, by comparison with
Table 1, were able to meet the requirements for the production phase scenario, where the maximum expected coverage range is less than 10 m.
Further tests were conducted with the aim of evaluating the coverage range of the sensors after they had been embedded into the produced artifacts, reflecting the final deployment phase. In these cases, the sensor module was directly surrounded by the building material and placed at higher distances than in the previous case, typically in outdoor locations.
For this experimental setup, the tests concerned only the Tinovi smart sensor, since the Microchip LoRa mote had already been assessed in a recent study [
34] under analogous conditions, and was demonstrated to be able to reach distances of more than 100 m when embedded into a sample artifact. Therefore, to perform the tests, the Tinovi sensors were embedded in a composite of river sand and binders like the final substrate used for the printing process.
Figure 6 illustrates two steps of this process relevant to the printing of nine cubes of 40 cm side.
Once the artifact was devised, the LoRa gateway was positioned inside the main ADMIN–4D office and the ED was placed in four different outdoor locations to evaluate its transmission behavior. As in the previous tests, the experiments were performed with the same spreading factor,
; transmission tests were considered as passed when the GW received at least three-quarters of the transmitted data without errors. The results are presented in
Figure 7, where the green lines stand for successful trials, while the red line indicates a failure in the transmissions.
The analysis of
Figure 7 highlighted that the Tinovi ED was able to quite efficiently covera range of 500 m in the absence of obstacles. The range was more limited in the case of strong obstructions in the line of sight path. Nevertheless, the final result was that the Tinovi ED was shown to be able to meet the coverage range requirements of the application.
5.2. Power Consumption and Battery Lifetime Estimation Tests
This subsection is concerned with the experimental activities devised to analyze the power consumption and battery lifetime of the selected LoRa EDs.
Neither the Microchip nor the Tinovi documentation provided the necessary specifications in terms of power consumption or absorbed current in the different working modes, i.e., either in sleep or idle state, and during transmission/reception. Nonetheless, as for both the EDs an indication about the transceiver model was available, it was possible to carry out a preliminary comparison, as shown in
Table 4.
A first set of measurements was then performed to evaluate the lifetime of sensors when they are equipped with low-cost, off-the-shelf, battery types. In this preliminary analysis, only the Microchip LoRa Mote was tested. This represents a type of worst case, since this mote cannot be fed with recent high energy density and capacity batteries, allowing the use only of classic AAA alkaline or rechargeable batteries. The tests were carried out using alkaline batteries from two different manufacturers [
35,
36]. The parameter “shelf life”, which represents an indication of the time the batteries will hold their charge without being used when stored under normal environmental conditions, was different for the two brands, being equal to five and ten years, respectively. This, however, represented a very low self discharge rate, and, considering the expected run time for the mote in this application, it may be argued that the final lifetime is almost exclusively dependent on the energy efficiency of the mote.
In the measurement campaign, the battery lifetime was estimated as the time elapsed between the instant the module was powered on and the timestamp of the last received LoRa packet. The Microchip LoRa mote does not provide any means to directly measure the charge status of the batteries, and, consequently, it was not possible to precisely track the discharge curve. Clearly, a significant parameter impacting on the lifetime is the sampling period required by the embedded sensor. Therefore, tests were performed in agreement with the values indicated in
Table 1. The results are provided in
Table 5.
As expected, the impact of the sending period on the battery lifetime was considerable. Moreover, batteries with longer shelf-life performed significantly better than their counterparts. Nevertheless, the increase in lifetime was not linearly dependent on increase in the sending period. This might be related to a lower energy efficiency of the mote when no transmission occurs, and specifically to a non-optimal management of the energy saving states available for the LoRa chip (i.e., switching between idle and sleep states) and the MCU.
Before presenting the next set of measurements, it is worth providing some observations regarding the available LoRa sensor nodes. From the assessments carried out to this point, the Microchip LoRa motes were not completely suitable for deployment in the context of the ADMIN 4D project. As already pointed out, such motes can not host battery types other than basic AAA batteries, preventing the assessment of more recent battery technologies. Moreover, these modules do not offer the possibility of remotely changing their transmission parameters, limiting the potential for fine tuning of the data rate and sending period. Finally, the Microchip motes require an enclosure size that, in some cases, may be larger than what is allowed by the application. Considering the worst case situation, i.e., the smallest possible artifact to be realized by the additive manufacturing application, the Microchip mote resulted in a sealed device which was unfortunately larger than the imposed bounds, since it needed a suitable external enclosure for both the mote and its external antenna [
34].
Therefore, in the next assessment, only the Tinovi LoRa ED was addressed since it demonstrated its compliance with the ADMIN 4D requirements. It can also be fed by more recent battery types characterized by more advanced chemistry than classic alkaline batteries. Moreover, it allows for a remote set-up, i.e., connection and transmission parameters can be dynamically changed even when the end devices are in movement or are physically unreachable. Additionally, the Tinovi sensor allows transmission of information about battery state of charge, ranging from 2.8 V to the maximum voltage reached during the charging (4.2 V) with a resolution of 1%. This feature is of particular importance for characterizing the behavior of batteries, as well as accurately comparing different battery typologies in terms of lifetime.
The following set of measurements are, therefore, devoted to a performance analysis of more recent high capacity batteries in this additive manufacturing scenario. Two different battery types were adopted: the first was an MKC 18,650 lithium-ion battery, while the second was a SAFT LS 17,500 lithium thionyl chloride battery. Their specifications are summarized in
Table 6. Both these technologies tend to became unstable (i.e., risk of explosion or fire) when used outside the operating temperature range, and this has to be carefully taken into account because the temperature reached inside the artifacts during the production phase may be quite high due to binder reaction processes. Nevertheless, a preliminary extensive analysis highlighted that, due to the IP67 enclosure, the battery operating temperatures always remained within the maximum allowed values.
The two different types of chemistry that were selected for this analysis are notably different, each possessing strengths and weaknesses. Lithium-ion (Li-ion) batteries have become the standard for consumer electronics. They are characterized by low cost, high availability and, most importantly, they can be recharged. However, in general, this type of battery suffers from a relatively high self-discharge rate and is strongly dependent on the discharge curve with temperature [
39].
These limitations do not apply to lithium thionyl chloride (LTC) batteries. These are primary cells, therefore not rechargeable, capable of delivering low currents, but characterized by both a flat discharge curve, in the range of the nominal voltage, and low dependence on the operating temperature [
40]. These characteristics make them particularly suitable for the ADMIN 4D project and, more generally, for supplying battery powered IoT measurement devices deployed in adverse environments.
In the following measurement approach, a transmission period of 5 min was adopted, representing a value typically adopted in the production phase of an artifact In the first set of measurements, Li-ion batteries were considered. A Tinovi ED was periodically sampled to obtain information about battery voltage, to derive the discharge curve. This is represented by the continuous blue line in
Figure 8.
From the figure, it can be observed that the employed Li-ion batteries ensured continuous and reliable operations for 18 days, after which the battery voltage dropped below the cutoff voltage. A further observation was that the obtained duration and discharge curve partially refuted the preliminary linear estimation model for Li-ion batteries proposed in [
41]. Thus, in order to achieve a better estimate of the battery lifetime for different temperature ranges, the popular model described in [
42,
43] was adopted (the same model is implemented in the Mathworks Simscape suite).
This model provides an estimate of the Li-ion battery voltage as
that is, the voltage is a (decreasing) function of the discharge current
i, as well as of the ambient and internal battery temperatures
T and
, respectively. The second term of Equation (
1),
, represents a discharge resistance dependent on the temperature equivalent to the thermistor model.
The function
contained in Equation (
1) contains several terms that can be derived from the analysis of the battery specifications. The model equation is
Some main terms can be identified in Equation (
2), corresponding to the different typical discharge phases of a battery. In particular, a temperature-dependent initial voltage
(fully charged) is followed by an exponential discharge phase, and by a nominal discharge phase. The description of the terms in Equation (
2) is reported below
is the voltage at the end of the exponential region, called the constant voltage ()
Q is the maximum battery capacity ()
K is the polarization constant (), often indicated as polarization resistance ()
i is the battery current ()
is a low-pass filtered version of the battery current, characteristic of this type of batteries, that often can be considered equal to i ()
is the actual battery charge ()
A is the exponential voltage ()
B is the exponential capacity ()
C is the nominal discharge curve slope ()
T is the cell or internal temperature ()
is the ambient temperature ()
A detailed analysis of the models, as well as an in-depth explanation of the different terms, and the way they can be obtained from the battery data sheet, can be found in [
42].
To obtain realistic results, the model parameters were tuned using the results of the experimental measurements carried out at 20
. Then the model was calibrated to match the boundaries of the discharge curve and to minimize the error in the central part. The probability density function of the calibration error is shown in
Figure 9. As can be seen, the calibration error was limited to the range −0.05
–0.04
, making it suitable for the intended use.
The calibrated model was then used to simulate the discharge curves for different temperature ranges. The results are reported in
Figure 8. As can be observed, the battery lifetime was strongly dependent on the operating temperature.
A second set of measurements was carried out considering the SAFT LS 17,500 LTC battery, using the same configuration and sending period for the previous experiment. The whole experimental campaign lasted about two months, and the results obtained are shown in
Figure 10.
The plot evidences the almost flat discharge curve characterizing the lithium thionyl chloride batteries. The batteries maintained a rather stable voltage of 3.56 V for nearly 40 days, a value very close to the 3.6 V nominal voltage. After this period, the voltage started to oscillate around 3.55 V. Then, after 50 days, the voltage commenced a quite abrupt decay and, in three to four days, reached the cut-off voltage where the sensor powered off.
As can be seen by comparing
Figure 8 and
Figure 10, the use of a LTC battery significantly increased the lifetime of the sensor.
It should be noted that the experiments with the two battery types were carried out with a sampling period of 5 min. This period is used in the production phase that typically lasts only a few hours, whereas in the final deployment phase, the sampling periods will be significantly longer. Thus, although a simple linear regression to predict the battery lifetime cannot be applied, it may be argued that, using a sampling period of the order of some hours (typical for the final deployment phase), the Tinovi sensor would be expected to last for a considerable amount of time (likely more than a year), which is a satisfactory lifetime for the ADMIN 4D project.