An active infiltration of information technology in the healthcare sector has led to a fundamental change in people's quality of life. In this regard, the security and safety problems of this technology using increase rapidly. This paper touches upon the issue of the healthcare Internet of Things (IoT) infrastructure failures of components and complete system. The purpose of the paper is to develop and research an availability model of a healthcare IoT system regarding failures of components. A detailed analysis of an architecture of healthcare IoT infrastructure is given. The main causes of the healthcare IoT based system failures are considered. Much attention is given to developing and research of a Markov model of a healthcare IoT system considering failures of components. Some essential high-level requirements that such system must meet are presented. The analysis of obtained simulation results showed the rates that have the greatest influence on the availability function of the healthcare IoT system.
1.1
The paradigm of the Internet of Things (IoT) implies the possibility of massively and
inexpensively connecting to an information network (for example, the Internet) any
physical object and control systems for these objects. IoT in general promises textually
to every citizen and every company, regardless of the industry - its own set of benefits
and improvements, savings and growth, the release of time and new opportunities. On
the basis of these statements, the IoT has already found applying in many industries.
According to predictive forecasts [
One of the most promising and already most advanced industries are medicine and healthcare. Networked medical and healthcare devices and their applications are already creating an Internet of Medical and/or Healthcare Things which is aimed at better health monitoring and preventive care for creating better conditions for patients who require constant medical supervision and/or preventive intervention. Healthcare and medical organizations (providers) also attempt to collect and analyze data that generate the IoT devices that are essential for prospective innovations.
One of the most sought-after fields in healthcare and medicine treatment, monitoring
and prognosis is Diabetes. According to [
But the new concepts and applying of new technologies bring certain risks including failures of devices, infrastructure which may lead to the worst outcome - the death of the user (patient). Hence to minimize such risks and assure required availability the system models and strategies of maintenance should be developed and researched. 1.2
For today there are a lot of papers that describe opportunities and benefits of using smart and intellectual technologies in the field of healthcare and medicine and at the same time they describe the security and safety problems of this technology using.
One of the most famous and almost all covering paper is [
The authors of [
Goševa-Popstojanova and Trivedi in [
The Markov model that takes into account the technical conditions of typical
network components of the IoT-based smart business center was presented in [
In [
The paper [
Nevertheless, despite a large number of researches regarding healthcare IoT, there are no papers that consider safety and reliability issues of healthcare IoT systems taking into account failures of hardware and software components and system failures. 1.3
The goal of the paper is to analyze and develop a model that describes the healthcare IoT system failures and their influence on availability indicators. Our approach is based on review of the variety of existing techniques and mathematical models for similar systems and step by step development of a set of states and transitions caused by failures of system components.
In this context, the paper proposes the Markov model describing possible failures of healthcare IoT system and recovery procedures. The remainder of this paper is organized as follows. The second section describes an architecture of healthcare IoT infrastructure and possible failures during its operation. The third is devoted to the development of the Markov model of a healthcare IoT system considering failures of components and analyzing simulation results. The last section concludes and discusses future research steps. 2 2.1
Analysis of the latest publications related to this topic [
In this paper, the main subject of the study is an insulin pump operating in the
infrastructure of the IoT. An activity diagram for the insulin pump operating
independently without interaction with other devices or Internet was described in
[
The data from the blood sugar sensor send to the blood sugar analysis and insulin requirement computation what is carried out by integrated technical possibilities and tools of the insulin pump and/or sends to the Cloud servers via the Internet gateway for further processing, storage, and visualization. The patient’s data can be analyzed using e.g. artificial intelligence tools in the Cloud. The decision made by artificial intelligence tools sends to the healthcare authority for the conclusive prescription and finally to the patient or insulin pump user. In more details, decisions that were made by the healthcare authority are also loaded into the Cloud, and then insulin pump user (control unit) downloads prescriptions. 2.2
It is clear that the healthcare IoT based system is a safety-critical system. If the pump or any other significant element fails to operate or does not operate correctly, then the patients’ health may be damaged or they may fall into a coma because their blood sugar levels are too high or too low, or the doctor’s prescription is not received by the patient, etc. There are, therefore, some essential high-level requirements that such system must meet: 1. The system shall be available to deliver insulin when required. 2. The system shall perform reliably and deliver the correct amount of insulin to counteract the current level of blood sugar. 3. Any component of the IoT system shall interact with any other when required. 4. The system shall be able to scale. 5. The Cloud component shall be able to process, storage and visualize all patients’ data when required. 6. The healthcare authority component shall be able to respond to all patient requests when required, etc.
Thereby as in any other information and technology systems, failures also may occur in the IoT based systems. Fig. 3 depicts in outline the main causes of healthcare IoT based system failures.
In papers [
In [
In general, in the healthcare IoT system, the failures of single subcomponents are possible. These failures may lead to the failures of the main components of infrastructure (i.e. insulin pump, cloud, etc.). In its turn, the failures of main components may lead to failure of the whole healthcare IoT system. Fig. 4 shows the dependence of the healthcare IoT system failures, where 0 – there is no any failure in the system, 1 – there is one failure (of subcomponent), 2 – there are two failures (subcomponent and main element), 3 – there are three failures (the failure of the whole healthcare IoT system).
In more details Fig. 5 shows a Markov graph of the functioning of the main components of healthcare IoT system during failures, λ - the failure rate, µ - the recovery rate. Thereby, the basic states of the healthcare IoT system are: 1 - normal condition (upstate) system; 2 - failure due to the power supply (battery) pump causes discharge, recharging and/or causing damage; 3 - failure of any one and/or more sensors of the insulin pump due to the out-of-order, does not deliver any output to inputs, delivers null output values and/or no meaningful values and/or impurity etc.); 4 - pump failure (inaccurate size/rate of insulin dose) due to the components defects, improper position of pump, ambient temperature, air pressure and/or design errors etc.; 5 - software of insulin pump control module failure due to buffer overflow or underflow, incorrect libraries, wrong algorithms or programming, threshold setting error etc.; 6 - hardware of insulin pump control module failure due to overheating, short or open circuit, high leakage current, high or low impedance, missed alarm, false alarm, fail to read/write data and/or design error etc.; 7 - intra wireless body area network (WBAN) communication failure due to the packet loss, isolation, a communication module failure (e.g., L2CAP, BNEP etc.), header corruption and/or length mismatch and/or payload corruption etc.; 8 - insulin pump (as the patient’s complex) failure due to the failure of any one or more main components; 9 - extra gateway communication partial failure due to data delivery failures; 10 - extra gateway communication partial failure due to Bluetooth/cellular/WiFi network unavailable; 11 - partial failure due to the refusal of the mobile application of the reader (control unit); 12 - cloud software failure due to planned or unplanned reboot, software updates and/or complex design; 13 - cloud hardware failure due to hard disk failures, RAID controller, memory and/or other devices; 14 - cloud scheduling failure due to overflow and/or timeout; 15 - cloud service failure due to request stage and/or execution stage; 16 - cloud failure due to power outage; 17 - cloud failure due to the failure of any one and/or more cloud components; 18 - failure due to incorrect assignment or programming of the device by a healthcare authority related to device functions or lack of functions; 19 - failure of the IoT healthcare system.
A system of Kolmogorov differential equations for presented Markov model is: +115 +116 +118)P1(t)+21P2 (t)+31P3 (t)+41P4 (t)+51P5 (t)+61P6 (t)+ +71P7 (t)+81P8 (t)+91P9 (t)+101P10 (t)+111P11(t)+121P12 (t)+131P13 (t)+ +141P14 (t)+151P15 (t)+161P16 (t)+171P17 (t)+181P18 (t)+191P19 (t); dP2 /dt = −21P2 (t)+12P1(t); dP3 /dt = −13P3(t)+13P1(t); dP4 / dt = −(41 +45)P4 (t)+14P1(t); dP5 /dt = −(51 +58)P6 (t)+16P1(t)+45P4 (t)+185P18 (t); dP6 / dt = −(61 +68)P6 (t)+16P1(t); dP7 / dt = −(71 +78)P7 (t)+17P1(t); dP8 /dt = −(81 +819)P8 (t)+58P5 (t)+68P6 (t)+78P7 (t); dP9 / dt = −9P9 (t)+19P1(t); dP10 /dt = −101P10 (t)+110P1(t); dP11 /dt = −111P11(t)+111P1(t)+1711P17 (t); dP12 /dt = −(121 +1217)P12 (t)+112P1(t); dP13 / dt = −(131 +1317)P13(t)+113P1(t);
Initial values are: dP17 / dt = −(1711 + 1719 + 171 )P17 (t ) + 1217 P12 (t ) + 1317 P13 (t ) + 1417 P14 (t ) +
P1(0) = 1, Pi(0) = 0, i = 2,3, …, 19.
To solve a system of the linear Kolmogorov differential equations it is necessary to carry out the collection and analysis of statistics on failures of healthcare IoT systems. 3.2
Hence the initial data for Markov model simulating were taken from [
The working state is state 1, and eighteen others are states with failures of different components and parts of the healthcare IoT system. The obtained probabilities of finding the healthcare IoT system in each state of Markov model are shown below (stationary values):
P1 = 0.9853745; P2 = 0.000103622; P3 = 0.003330566; P4= 0.000251795; P5 = 0.001162896; P6 = 0.0003859747; P7 = 0.006145591; P8 = 0.0009757008; P9 = 0.001486934; P10 = 0.0006328081; P11 = 3.207395e-05; P12 = 3.070724e-05; P13 = 1.056492e-05; P14 = 2.90451e-05; P15 = 2.596815e-05; P16 = 5.734457e-06; P17 = 3.038937e-08; P18 = 1.545724e-05; P19 =2.957985e-08.
Hence A(t) = P1(t). Fig. 6 shows the availability function changing before a transition to the stationary value (Astationary = 0.9853745). According to the simulation results the function gets a qua approximately at step 2300 h, i.e. 3 months later after beginning of work. Fig. 7 – 10 show the dependence of the availability function changing depending on the different types of failures changing on the healthcare IoT systems rates. Fig. 8. Dependence of the availability function changing depending on the changing λ15 rate. Fig. 9. Dependence of the availability function changing depending on the changing λ110 rate. The obtained results analysis shows that the greatest influence on the change in the availability function is the λ15 rate and the next is the λ13 rate (i.e. different components of the patient device (for our case of insulin pump) failures). The least influence has the failures of cloud components due to the rapid recovery time. These results are confirmed by statistical data.
The analysis of obtained results shows that the complete failure of the healthcare IoT system does not happen too often (one case on the analyzed time interval due to the complete failure of the Cloud). Nevertheless, failures of constituent elements of the system arise quite often that may affect the performance of mission-critical functions of the healthcare IoT system and in the worst case, lead to the death of the patient. The most often failures are due to the failure of the insulin pump and its particular elements and components and some components of the Cloud.
Availability of the system can be improved by more fast recovery (repair) of the equipment and system resources and application of more reliable devices. 4
Due to the use of the IoT technologies, the interaction of objects, environment, and people will be extremely active, and it is making it possible to hope that the world will be "smart" and a well-appointed for a person. However, at the same time, the IoT faces a number of problems that can prevent us from taking power of its potential advantages.
In this paper, the overview of the healthcare IoT system failures is presented. Based on the conducted analysis and classification of the main possible failures of healthcare IoT infrastructure a Markov model considering failures of components is constructed. For the developed model probabilities of finding IoT system in each state of Markov model are shown. The obtained results show possible most frequent failures of healthcare IoT system. The presented Markov model can be used not only for the availability evaluation of insulin pumps but for other healthcare devices operating in IoT system.
Next steps of research will be dedicated to a development of more general dependability models for healthcare IoT systems and combining results of this paper and models taking into account both the reliability, safety and security requirements and issues.
This paper implies results obtained during involvement in the Erasmus+ programme educational project ALIOT «Internet of Things: Emerging Curriculum for Industry and Human Applications» (reference number 573818-EPP-1-2016-1-UK-EPPKA2-CBHEJP, web-site http://aliot.eu.org) in which the appropriate course is under development (ITM4 - IoT for health systems). Within its framework, the teaching modules related to IoT systems modelling were developed. The authors would like to thank colleagues on this project, within the framework of which the results of this work were discussed.
The authors also would like to show deep gratitude to colleagues from Department of Computer Systems, Networks and Cybersecurity of National Aerospace University n. a. N. E. Zhukovsky «KhAI» for their patient guidance, enthusiastic encouragement and useful critiques of this paper.